Yes — with 23.7 GB to spare
Gemma 3 4B at Q4_K_M fits your M3 Pro · 36 GB entirely in unified memory at 8K context, at an estimated 35 tokens per second. There is room for its full 128K window.
Fully in unified memory
8K context
Q4_K_M · 2.4 GB
Gemma Terms of Use
Released Mar 2025
Vision
Vision-capable at 4B. Superseded by Gemma 4 E4B, still everywhere.
The VRAM budget
weights 2.4 GB
Weights 2.4 GB
KV cache @ 8K 0.30 GB
Runtime overhead 0.6 GB
Free 23.7 GB of 27.0 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 8.0 GB | 8.9 GB | 128K | 10 | Reference | Long context |
| Q8_0 | 4.3 GB | 5.2 GB | 128K | 20 | −0.1% ppl | Long context |
| Q6_K | 3.3 GB | 4.2 GB | 128K | 26 | −0.4% ppl | Long context |
| Q5_K_M | 2.8 GB | 3.7 GB | 128K | 30 | −0.8% ppl | Long context |
| Q4_K_M | 2.4 GB | 3.3 GB | 128K | 35 | −1.9% ppl | Recommended |
| Q3_K_M | 2.0 GB | 2.9 GB | 128K | 43 | −5.4% ppl | Long context |
| Q2_K | 1.7 GB | 2.6 GB | 128K | 50 | −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. This model interleaves sliding-window layers (1024 tokens, 1 global in 6), which is why its cache barely grows with context.
How to run it
$ pip install mlx-lm $ mlx_lm.generate --model mlx-community/gemma-3-4b-it-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.
01Download is 2.4 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 27.0 GB of 36 GB. Raising it with iogpu.wired_limit_mb is possible, and risky.
03There is room to go to the model's full 128K context on this card.