Home › Models › Llama 3.2 3B Instruct › Apple M4 Max (40-core GPU)
Llama 3.2 3B Instruct on the Apple M4 Max (40-core GPU)
Llama 3.2 3B Instruct runs at about 208 tokens/s on the Apple M4 Max (40-core GPU) (measured, MLX-4bit, MLX, apple-silicon-llm-bench).
MLX-4bit · MLX
The source gives no prompt-processing figure.
Every run
| Quant | Backend | Generate tok/s | Prompt tok/s | Test | Build / by / notes | Source |
|---|---|---|---|---|---|---|
| MLX-4bit | MLX | 208 | – | short-chat decode (warm median) | john-rocky mlx-swift, mlx-community/Llama-3.2-3B-Instruct-4bit (Q4); Mac Studio M4 Max 40-core GPU; captured 2026-07-13; warm = median of same-session warm runs, thermal-nominal start | john-rocky apple-silicon-llm-bench (MIT) |
tg128 = generating 128 tokens; pp512 = reading a 512-token prompt (llama-bench). LocalScore averages several prompt and output lengths, so its numbers run lower than tg128. "decode, llama-server" = single-stream generation speed measured against a running llama-server (no speculative decoding).
No card yet?
Try it on a rented GPU (RunPod)Affiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.
All results for the Apple M4 Max (40-core GPU) · Llama 3.2 3B Instruct on other hardware
Embed this result
HTML
<a href="https://rigbench.pages.dev/model/llama-3-2-3b-instruct/on/apple-m4-max-40c/"><img src="https://rigbench.pages.dev/badge/llama-3-2-3b-instruct--apple-m4-max-40c.svg" alt="Llama 3.2 3B Instruct · Apple M4 Max (40-core GPU): 208 tok/s (MLX-4bit, measured)"></a>
Markdown (GitHub README, forums)
[](https://rigbench.pages.dev/model/llama-3-2-3b-instruct/on/apple-m4-max-40c/)