The M5 Ultra Mac Studio runs local models at 1228 GB/s of memory bandwidth with 96 to 512 GB of unified memory. On Apple Silicon that memory is shared with the GPU, so the whole pool is available for weights: at 512 GB you can hold roughly a 833B dense model at Q4. Two Max dies fused together: the widest memory bus and the highest capacity Apple sells. This is the tier that runs frontier-size open weights locally.
Memory bandwidth is faster than 95% of the Apple Silicon chips shipped in a Mac, against a 1228 GB/s peak.
Its memory ceiling is above 90% of them, against a 512 GB peak.
Every option Apple sells with this chip. The model list below recomputes against the one you pick.
GPU cores
Unified memory
Apple couples memory to the core count on this chip, so the options change with the bin above.
Unified memory is the ceiling and it is soldered, so this is the decision you cannot revisit.
6,418 of 6,563 models fit by their known memory.
A browser cannot measure a runtime's working memory, so no fit here is confirmed. ToolPiper measures the file on your Mac and confirms it.
6,499 of 6,563 models fit by their known memory.
A browser cannot measure a runtime's working memory, so no fit here is confirmed. ToolPiper measures the file on your Mac and confirms it.
6,550 of 6,563 models fit by their known memory.
A browser cannot measure a runtime's working memory, so no fit here is confirmed. ToolPiper measures the file on your Mac and confirms it.
Every model in the database against this exact configuration, at 1228 GB/s. Ratings and speeds are the same numbers the model pages show.
Showing 6563 of 6563 models
General · goekdeniz-guelmez · 2026-07-31
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · goekdeniz-guelmez · 2026-07-31
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · internscience · 2026-07-13
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2026-07-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2026-06-24
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · ma7ee7 · 2026-07-30
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · nanbeige · 2026-07-21
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · radixark · 2026-07-27
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · weiboai · 2026-06-12
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · deepreinforce-ai · 2026-06-21
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · empero-ai · 2026-06-19
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · ktruestory · 2026-05-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · obliteratus · 2026-06-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · huihui-ai · 2026-07-11
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · openbmb · 2026-05-21
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · yuxinlu1 · 2026-06-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2026-03-31
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · lukebailey181pub · 2026-04-21
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · obliteratus · 2026-04-15
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · google · 2026-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · ibm-granite · 2026-04-06
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · ibm-granite · 2026-04-06
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · ibm-granite · 2026-04-16
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · ibm-granite · 2026-04-16
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Reasoning · jackrong · 2026-03-16
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Reasoning · openonerec · 2026-06-09
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2026-02-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2026-02-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2026-02-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2026-02-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2026-02-27
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2026-02-27
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2026-02-27
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2026-02-26
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · cagrigungor · 2026-08-06
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Google · 2026-03-02
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Google · 2026-03-02
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · nvidia · 2026-03-02
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · nvidia · 2026-03-18
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · pearl-ai · 2026-02-26
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · davidau · 2026-02-02
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2026-01-20
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · ravichandranj · 2026-02-13
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · alibaba-nlp · 2026-03-31
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · dreamfast · 2026-01-11
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · frontiersmind · 2026-08-03
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-12-25
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2026-01-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2026-01-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · Liquid AI · 2026-01-06
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2026-01-04
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Liquid AI · 2026-01-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-10-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Liquid AI · 2025-10-22
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · xcuros · 2026-02-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · allenai · 2025-11-19
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · allenai · 2025-11-17
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · allenai · 2025-11-18
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · allenai · 2026-01-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-09-22
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-09-30
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · openonerec · 2025-12-30
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · allenai · 2025-09-12
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · anton-hugging · 2026-02-06
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · baseten · 2025-09-12
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · ibm-granite · 2025-09-16
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · ibm-granite · 2025-09-16
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · typhoon-ai · 2025-09-23
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · utter-project · 2026-01-26
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · z-lab · 2026-01-04
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · z-lab · 2026-01-04
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · bytedance · 2025-10-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · lgai-exaone · 2025-07-11
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-07-10
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-08-22
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-09-03
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-09-03
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-07-10
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-09-03
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-09-03
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-08-25
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Liquid AI · 2025-07-10
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Liquid AI · 2025-08-12
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Liquid AI · 2025-08-12
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · Alibaba · 2025-08-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Alibaba · 2025-08-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · hmellor · 2025-07-22
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · nvidia · 2025-08-21
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · nvidia · 2025-08-12
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · swiss-ai · 2025-08-13
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · t-tech · 2025-12-22
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · uzlm · 2025-09-03
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · avitotech · 2025-10-20
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · menlo · 2025-06-25
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2026-08-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · amd · 2025-05-17
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Reasoning · DeepSeek · 2025-05-29
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · droplychee · 2026-07-17
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · stefanruseti · 2025-06-04
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · bananamind · 2026-07-17
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · dream-org · 2025-04-03
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · dream-org · 2025-04-03
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · dreamfast · 2026-04-30
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · huggingfacetb · 2025-07-08
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · lgai-exaone · 2025-03-12
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · maliosdark · 2026-07-09
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Alibaba · 2025-09-23
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Alibaba · 2025-09-23
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · Alibaba · 2025-09-23
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · shaungves · 2026-06-19
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · sunbird · 2026-07-19
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · viorikaai-org · 2026-07-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · fableforge-ai · 2026-07-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · hahayhe · 2026-06-25
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · ibm-granite · 2025-04-09
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · ibm-granite · 2025-04-09
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · ibm-granite · 2025-10-07
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · k-chirkunov · 2026-07-14
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Coding · mahiatlinux · 2026-06-30
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Reasoning · Microsoft · 2025-04-29
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · mlabonne · 2025-05-28
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · naver-hyperclovax · 2025-07-21
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · naver-hyperclovax · 2025-04-22
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · neuralcrew · 2025-09-14
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · nvidia · 2025-03-16
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · openbmb · 2025-09-02
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · raidium · 2026-06-15
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Embedding · taide · 2026-06-12
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · z-lab · 2026-04-23
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · huggingfacetb · 2025-06-19
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · Alibaba · 2025-01-23
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2025-01-26
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2025-01-26
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Multimodal · Alibaba · 2026-04-21
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · darthcrawl · 2026-05-07
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Reasoning · DeepSeek · 2025-01-20
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Reasoning · DeepSeek · 2025-01-20
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Reasoning · DeepSeek · 2025-01-20
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · internlm · 2025-01-13
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · Microsoft · 2025-02-19
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · Microsoft · 2025-02-24
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · pfnet · 2025-02-05
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · preparebuddy · 2026-06-02
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · sapientinc · 2026-05-17
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · z-lab · 2026-03-14
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
General · astromlab · 2024-11-14
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Reasoning · jackrong · 2026-02-27
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · k-intelligence · 2025-07-03
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · lgai-exaone · 2024-12-01
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Chat · lgai-exaone · 2024-12-01
Unvalidated forecasts for an 8K text context under a declared full-GPU scenario, at the lowest GPU bin sold with the chosen memory. Memory is reported as its components: exact artifact bytes and a KV allocation derived from a declared attention layout where the catalog carries them, a parameter proxy with no error bound where it does not, and the runtime's working memory, which a browser cannot measure. A fit is confirmed only when every component is known, so a model whose known memory fits reads "Fit unconfirmed" until ToolPiper measures it on your Mac, and one whose known memory alone exceeds the budget reads "Won't fit". Estimated speed ignores prompt length and prefill, and can overstate long-context performance. No measured speed error range is available. The recommendation score weighs declared preferences, not measured answer quality, and becomes a range where evidence is missing.
Known memory fits; the runtime's working memory is not measured.
Both bins run the same 1228 GB/s memory bus, so token generation is the same on either one. The extra cores show up in image and video work, not in tokens per second.
| Configuration | Memory bandwidth | Memory options | Models that fit |
|---|---|---|---|
| 30-core CPU, 64-core GPU | 1228 GB/s | 96 GB | Identical |
| 36-core CPU, 80-core GPU | 1228 GB/s | 256, 512 GB | Identical |
No compatible observations are published here. Formula estimates remain labelled, and unsupported forecasts remain unavailable.
Benchmark contribution is disclosed before a run, and the leaderboard shows every chip that already has one.
What each step actually changes for local models, rather than which one is newer.
Apple has not announced any of this. The M7 Max and Ultra are the first parts Apple designed after cancelling a generation to reach them, so extrapolating from the M5 under-represents them. These rows assume LPDDR6, whose wider channels grow every bus by half, at its top speed bin by the time the Max and Ultra ship. The 1.5 TB Ultra ceiling is Bloomberg's reported design target, and whether that configuration ships depends on the memory market. Stacked memory or a new package fabric would land above these numbers; nobody outside Apple can price that yet.
Projected chip · expected 2028
2765 GB/s · 256 to 1536 GB unified memory
40-core CPU · 112 or 128-core GPU · 152 TOPS Neural Engine
Would hold about a 2501B model at Q4
On a PC the model has to fit in GPU VRAM, which is a separate pool from system RAM and usually the smaller of the two. Apple Silicon has one pool. The M5 Ultra's 1228 GB/s bus is shared by CPU, GPU, and Neural Engine, so a 512 GB machine can hand almost all of that to a model with no copy across a bus.
The Studio exists for this workload. It carries the widest memory buses and the highest capacities Apple sells, and it runs at full clocks indefinitely. Buy the memory, not the cores: every extra GB raises what you can load, while the core count only moves throughput on models that already fit.
Yes, at 512 GB. A 70B model at Q4_K_M needs about 38 GB including an 8K context, and 512 GB of unified memory leaves about 450 GB for weights once macOS takes its share. At 96 GB it does not fit at any quantization worth running.
Memory is the only spec that changes what you can run at all. 96 GB holds about a 155B model at Q4; 512 GB holds about 833B. It is soldered, so this is a one-time decision, and it is the upgrade worth paying for before core count.
Token generation is bandwidth-bound, so M5 Ultra throughput scales with its 1228 GB/s memory bus. Divide bandwidth by the size of the weights actually read per token to get the ceiling, then expect roughly half of that in practice. A 7B model at Q4 reads about 4 GB per token pass, so the M5 Ultra lands in the tens of tokens per second and a 70B model lands in the single digits.
Not for LLMs. Both bins run the same 1228 GB/s memory bus and take the same memory options, and token generation is bound by bandwidth rather than GPU cores. The extra cores show up in image generation and video work, not in tokens per second.
Buy on the memory you need today. Apple raises memory ceilings slowly and bandwidth in steps, and the M5 Ultra already holds about a 833B model at Q4. If your target model fits in 512 GB, waiting buys throughput rather than capability.
ToolPiper downloads, manages, and runs local models on Apple Silicon. Free, and nothing leaves the machine.