The M1 MacBook Air runs local models at 68.25 GB/s of memory bandwidth with 8 to 16 GB of unified memory. On Apple Silicon that memory is shared with the GPU, so the whole pool is available for weights: at 16 GB you can hold roughly a 20B dense model at Q4. A base M chip is the narrow end of the memory bus. It runs small models pleasantly and stops hard at the memory ceiling, which is the constraint you will hit first.
Apple no longer sells this configuration new. It stays fully evaluated here because the used market is where most of its local AI value now sits.
Memory bandwidth is faster than 0% of the Apple Silicon chips shipped in a Mac, against a 1228 GB/s peak.
Its memory ceiling is above 0% 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
Unified memory is the ceiling and it is soldered, so this is the decision you cannot revisit.
4,997 of 6,563 models fit, and 3,331 of them run with headroom rather than as a squeeze.
5,572 of 6,563 models fit, and 5,054 of them run with headroom rather than as a squeeze.
Every model in the database against this exact configuration, at 68.25 GB/s. Ratings and speeds are the same numbers the model pages show.
Showing 6563 of 6563 models
General · Liquid AI · 2026-06-24
General · Liquid AI · 2026-03-31
General · cagrigungor · 2026-08-06
Multimodal · Alibaba · 2026-02-28
General · openbmb · 2026-05-21
Multimodal · Alibaba · 2026-02-28
Reasoning · openonerec · 2026-06-09
General · ktruestory · 2026-05-28
General · Liquid AI · 2025-10-28
General · frontiersmind · 2026-08-03
Multimodal · Liquid AI · 2025-08-12
General · Liquid AI · 2026-01-20
General · Liquid AI · 2025-09-30
General · Liquid AI · 2025-09-03
General · Liquid AI · 2025-08-25
General · Liquid AI · 2025-09-03
General · petrouil · 2026-07-23
General · Liquid AI · 2026-01-05
General · Liquid AI · 2025-07-10
General · Liquid AI · 2026-01-04
Chat · Liquid AI · 2026-01-06
General · ibm-granite · 2025-10-07
General · Liquid AI · 2025-07-10
General · viorikaai-org · 2026-07-05
General · bananamind · 2026-07-17
General · maliosdark · 2026-07-09
General · raidium · 2026-06-15
Embedding · taide · 2026-06-12
Multimodal · datalab-to
General · Alibaba · 2025-04-28
General · distil-labs
General · Alibaba · 2025-09-23
General · Liquid AI · 2025-07-10
Multimodal · Liquid AI · 2026-01-05
Multimodal · Liquid AI
General · Liquid AI
General · onnx-community · 2025-04-28
General · baidu
General · openbmb · 2025-06-05
General · farbodtavakkoli
General · kamilamila
General · ordenwills
General · Microsoft
General · Microsoft
General · tencent
General · primeintellect
General · Liquid AI · 2025-08-22
General · Liquid AI · 2025-09-03
General · Liquid AI · 2025-09-03
General · darthcrawl · 2026-05-07
General · appvoid
General · pinkstack
Coding · Liquid AI
General · squ11z1
General · iselabvn
General · meddies
General · mihaipopa-1
General · lazos
General · q1ngmang
General · baidu
Multimodal · Liquid AI
General · TII
Reasoning · supralabs
General · kylesayrs
General · inference-optimization
General · Alibaba · 2025-04-27
General · Alibaba · 2024-05-31
General · hmellor · 2025-07-22
Multimodal · tencent
Multimodal · ath-maas
General · Google · 2022-03-02
Multimodal · paddlepaddle
General · stefanruseti · 2025-06-04
Multimodal · paddlepaddle
General · etherll
Multimodal · paddlepaddle
Reasoning · jackrong
Coding · rahul7star
General · limloop
General · melikegks
Coding · dalatexcoder
General · iselabvn
Multimodal · 8f-ai
General · cooperdk
Multimodal · dingdust
Multimodal · yuandaxia
Reasoning · healshsj
Multimodal · inclusionai
Multimodal · Alibaba · 2026-02-28
Multimodal · Alibaba · 2026-02-28
General · Microsoft
General · Microsoft
General · farbodtavakkoli
General · voyageai
General · farbodtavakkoli
General · TII · 2025-05-01
General · jakobhuss
General · sapientinc · 2026-05-17
Coding · shahriarferdoush
General · Liquid AI
General · openbmb
General · mesolitica
General · benjamin
Multimodal · Liquid AI · 2025-08-12
General · numind
General · divelab
General · toxicityprompts
General · lgai-exaone · 2025-07-11
General · ibm-granite
General · saidutta69
General · kiddiszc
General · lmms-lab
General · tabularisai
General · numind
General · isotonic
General · manycore-research
General · bezzam
General · ibm-granite
General · huihui-ai
General · skis-ai-research
General · thkim0305
General · lunahr
General · ibm-granite
General · lunahr
General · acon96
General · reaperdoesntknow
General · kerzgrr
General · kerzgrr
General · gokaygokay
General · litert-community
General · rishiraj
General · multiversecomputingcai
General · ibm-granite
General · etherll
General · reallexi
General · n99av80n
General · reallexi
General · davidau
General · hamoai
General · quipuai
General · occ-ai
General · p-e-w
General · kulpreetkaurs
General · dingdust
Both bins run the same 68.25 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 |
|---|---|---|---|
| 8-core CPU, 7-core GPU | 68.25 GB/s | 8, 16 GB | Identical |
| 8-core CPU, 8-core GPU | 68.25 GB/s | 8, 16 GB | Identical |
Nobody has submitted a benchmark on the M1 yet, so every speed on this page is the formula estimate rather than a measured run. The estimate is bandwidth-driven and calibrated against chips that do have data, which makes it a good guide and not a promise.
ToolPiper contributes a result anonymously when you run the benchmark, 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.
Shipping chip, not yet in this Mac · expected 2026
153 to 170 GB/s · 16 to 32 GB unified memory
12-core CPU · 12-core GPU · 76 TOPS Neural Engine
Would hold about a 42B model at Q4
Projected chip · expected 2027
240 GB/s · 16 to 48 GB unified memory
12-core CPU · 12 or 16-core GPU · 76 TOPS Neural Engine
Would hold about a 63B 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 M1's 68.25 GB/s bus is shared by CPU, GPU, and Neural Engine, so a 16 GB machine can hand almost all of that to a model with no copy across a bus.
Apple stopped selling this one, which is exactly why it is interesting. The Air has no fan, so a long generation run settles into a lower sustained clock than the same chip in a Pro. Model loading and short chats are unaffected; a multi-hour batch job is where you would notice. A used M1 at 16 GB still gives you 68.25 GB/s and a hard 20B ceiling, and neither number degrades with age the way a battery does.
No. A 70B model at Q4_K_M needs about 46 GB, and the largest M1 MacBook Air tops out at 16 GB, which leaves about 14 GB for weights. The practical ceiling on this machine is around 20B parameters at Q4.
Memory is the only spec that changes what you can run at all. 8 GB holds about a 9B model at Q4; 16 GB holds about 20B. 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 M1 throughput scales with its 68.25 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 M1 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 68.25 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.
For inference, the specs that matter do not age: 68.25 GB/s and up to 16 GB of unified memory are the same numbers today as they were in 2020. A used M1 at the top memory option usually beats a new base-tier machine at the same price on both. Check the battery and the display, not the silicon.
ToolPiper downloads, manages, and runs local models on Apple Silicon. Free, and nothing leaves the machine.