What is the largest LLM the Apple M3 Pro (36GB) can run?

Qwen3.6 35B-A3B, at Q4_K_M using about 22.6 GB of the 24.1 GB available, at roughly 49 tokens per second (8K context).

How many local AI models run well on the Apple M3 Pro (36GB)?

48 of the 69 models we track run fully in its 36 GB of unified memory at Q4_K_M or better.

Generation speed follows memory bandwidth. With 150 GB/s, expect about 6 tokens per second on an 8B model.

Every model we track, checked against the Apple M3 Pro (36GB) at 8K context. The largest that runs well is Qwen3.6 35B-A3B.

Unified memory36 GB24.1 GB usable by the GPU

Bandwidth150GB/s

Runs well48/69models at Q4_K_M or better

Fastest model~60tok/s · Qwen3 0.6B

Bars show memory used out of 24.1 GB. Pick a row for the full breakdown.

NameMemory usedBest quantSpeed

NameMemory usedBest quantSpeed

NameMemory usedBest quantSpeed

Run it in the cloud

Rent a cloud GPU by the hour. No upfront cost, stop any time.