M2 Max LLM benchmark
No signed M2 Max submissions yet. Check the configuration guidance below, or contribute a reproducible measurement from your own machine.
Specs vendor specs, not measured
- Unified memory (max configuration)
- 96 GB
Plan memory before choosing a model
The catalog lists a family maximum of 96 GB of unified memory. Enter your installed amount; smaller configurations have less space. Model weights, context cache and runtime overhead must share the available memory. Capacity alone does not establish which model will fit or how fast it will run. Use the memory-fit calculator for your model and context as an estimate, then verify with the intended runtime. No signed M2 Max measurement is available yet.
Contribute a signed M2 Max benchmark
Run on YOUR hardware to populate this page: pipx install llm-speed && llm-speed bench
$ pipx install llm-speed && llm-speed bench
Community folklore on M2 Max
9 unverified claims extracted from Reddit/HN comments. Lower trust than signed runs above; every row links to the source.
- communityconfidence 70%
27.00tok/s — Qwen2.5-coder-14B on M2 Max via mlx
our signed data: M2 Max · Qwen2.5-coder-14B
“ith Qwen2.5-coder-32B Q4 MLX at 11t/s with a context size of 12K while using Xcode, Fork, Safari and Music. I use Qwen2.5-coder-14B MLX Q4 (27t/s) with full context size when I need longer conversation. To me, 32B Q4 is the best ratio quality/speed we can get on our machines.”
- communityconfidence 70%
11.00tok/s — Qwen2.5-coder-32B on M2 Max via mlx
our signed data: M2 Max · Qwen2.5-coder-32B
“I have the 32GB M2 Max and am currently working with Qwen2.5-coder-32B Q4 MLX at 11t/s with a context size of 12K while using Xcode, Fork, Safari and Music. I use Qwen2.5-coder-14B MLX Q4 (27t/s) with full context size when I”
- communityconfidence 60%
50.00tok/s — Qwen 3 30B A3B on M2 Max via mlx
our signed data: M2 Max · Qwen 3 30B A3B
“I have an M2 Max. With Qwen 3 30B A3B: * GGUF Q4KM = 50Tps * GGUF Q8 = 38Tps * MLX Q4KM = 70Tps * MLX Q8 = 50 Tps I am currently experimenting with using LM Studio to run my backend and using via Open WebUI for MLXQ8.”
- communityconfidence 60%
70.00tok/s — Qwen 3 30B A3B on M2 Max via mlx
our signed data: M2 Max · Qwen 3 30B A3B
“I have an M2 Max. With Qwen 3 30B A3B: * GGUF Q4KM = 50Tps * GGUF Q8 = 38Tps * MLX Q4KM = 70Tps * MLX Q8 = 50 Tps I am currently experimenting with using LM Studio to run my backend and using via Open WebUI for MLXQ8.”
- communityconfidence 60%
38.00tok/s — Qwen 3 30B A3B on M2 Max via mlx
our signed data: M2 Max · Qwen 3 30B A3B
“I have an M2 Max. With Qwen 3 30B A3B: * GGUF Q4KM = 50Tps * GGUF Q8 = 38Tps * MLX Q4KM = 70Tps * MLX Q8 = 50 Tps I am currently experimenting with using LM Studio to run my backend and using via Open WebUI for MLXQ”
- communityconfidence 55%
18.00tok/s — on M2 Max via mlx Q4_K_M
our signed data: M2 Max
“That's a little bit surprising. I'm a daily user of Devstral-Small-2 24B 4Bit MLX on an M2 Max MBP (32GB 400GB/s) and I get 18t/s on average. I use it extensively as an agent in VSCode and Xcode, so no small tasks nor small context. I understand that MLX is faster than”
- communityconfidence 50%
20.00tok/s — on M2 Max via llama.cpp
our signed data: M2 Max
“ke there are a couple of issues:~~ - ~~first, there is no real performance difference compare to the non REAP version, I get pp 328 t/s tg 20 t/s for this model, while the non REAP was pp 320 t/s tg 20 t/s (same tg, slightly faster pp for REAP)~~ - ~~memory usage was the only go…”
- communityconfidence 50%
10.00tok/s — on M2 Max via llama.cpp
our signed data: M2 Max
“this REAP model. Update: I just run again on my 5060 machine, really impressed, while the non-REAP model gives me at most pp 80 t/s, tg 10 t/s for 200k context window, this REAP model reached pp 330 t/s, tg 20 t/s. --- Old comment: ~~I ran some test on a M2 Max MBP, and loo…”
- communityconfidence 50%
80.00tok/s — on M2 Max via llama.cpp
our signed data: M2 Max
“t to create this REAP model. Update: I just run again on my 5060 machine, really impressed, while the non-REAP model gives me at most pp 80 t/s, tg 10 t/s for 200k context window, this REAP model reached pp 330 t/s, tg 20 t/s. --- Old comment: ~~I ran some test on a M2 Max …”
Request a run on M2 Max
No measured run yet for these pairs. Vote for the one you want; the most-requested runs go first. See all requests.
We never sell or share your email. Privacy