No — not on this device
Gemma 4 31B at Q4_K_M needs 20.2 GB against 10.7 GB usable, and the shortfall of 9.5 GB is more than 128 GB of system RAM can cover at a tolerable speed. A smaller sibling or a lower quantisation is the honest answer here.
The dense flagship: strongest maths of the 24–32 GB class (89% AIME), clean prose, vision. Q4 is a tight 24 GB fit.
What hardware do I need for Gemma 4 31B? →
Fits instead: Gemma 4 12B (8.2 GB) · Gemma 4 E4B (5.2 GB)
The VRAM budget
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 31.0 GB | 33.6 GB | — | ~1.2 | −0.1% ppl | 22.9 GB over |
| Q6_K | 23.9 GB | 26.5 GB | — | ~1.6 | −0.4% ppl | 15.8 GB over |
| Q5_K_M | 20.7 GB | 23.3 GB | — | ~1.8 | −0.8% ppl | 12.6 GB over |
| Q4_K_M | 17.6 GB | 20.2 GB | — | ~2.2 | −1.9% ppl | 9.5 GB over |
| Q3_K_M | 14.2 GB | 16.9 GB | — | ~2.7 | −5.4% ppl | 6.2 GB over |
| Q2_K | 12.2 GB | 14.8 GB | — | ~3.1 | −15% ppl | 4.1 GB over |
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, 10 of 60 layers global), which is why its cache barely grows with context.
How to run it
$ pip install mlx-lm $ mlx_lm.generate --model mlx-community/gemma-4-31B-it-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.