Yes — with 20.4 GB to spare
Ministral 3 3B at Q4_K_M fits your M1 Pro · 32 GB entirely in unified memory at 8K context, at an estimated 52 tokens per second. Past 209K the KV cache pushes it over — quantise the cache to q8_0, or step down a quantisation, to go longer.
Fully in unified memory
8K context
Q4_K_M · 2.2 GB
Apache 2.0
Released Dec 2025
Vision
Edge model with a vision encoder and a 256K window. Apache 2.0.
The VRAM budget
weights 2.2 GB
Weights 2.2 GB
KV cache @ 8K 0.81 GB
Runtime overhead 0.6 GB
Free 20.4 GB of 24.0 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 7.2 GB | 8.6 GB | 159K | 16 | Reference | Long context |
| Q8_0 | 3.8 GB | 5.2 GB | 192K | 29 | −0.1% ppl | Long context |
| Q6_K | 2.9 GB | 4.4 GB | 201K | 38 | −0.4% ppl | Long context |
| Q5_K_M | 2.5 GB | 4.0 GB | 205K | 44 | −0.8% ppl | Long context |
| Q4_K_M | 2.2 GB | 3.6 GB | 209K | 52 | −1.9% ppl | Recommended |
| Q3_K_M | 1.8 GB | 3.2 GB | 213K | 64 | −5.4% ppl | Long context |
| Q2_K | 1.5 GB | 2.9 GB | 215K | 74 | −15% ppl | Long context |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it.
How to run it
$ pip install mlx-lm $ mlx_lm.generate --model mlx-community/Ministral-3-3B-Instruct-2512-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.
01Download is 2.2 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02macOS caps what the GPU may wire down at about 24.0 GB of 32 GB. Raising it with iogpu.wired_limit_mb is possible, and risky.
03There is room to go to 209K context on this card.