Real measured prompt-processing and token-generation speed for local LLMs on Apple Silicon, contributed anonymously from ToolPiper. Each score is the geometric mean of pp512, tg128, and time-to-first-token, recomputed server-side — comparable to a LocalScore number. Same workload label compares only to the same label.
Updated 2026-06-16 · 0 measured results
The leaderboard fills in as people run the benchmark in ToolPiper. Until then, the model fit checker shows formula-based speed estimates for every Mac.