Yes — with 63.8 GB to spare
Gemma 4 12B at Q4_K_M fits your M2 Max · 96 GB entirely in unified memory at 8K context, at an estimated 33 tokens per second. There is room for its full 256K window.
Fully in unified memory
8K context
Q4_K_M · 6.7 GB
Apache 2.0
Released 29 May 2026
Vision
The "unified" Gemma 4: text, image and audio in one 12B that fits a 12 GB card at Q4. 140+ languages.
The VRAM budget
weights 6.7 GB
Weights 6.7 GB
KV cache @ 8K 0.81 GB
Runtime overhead 0.6 GB
Free 63.8 GB of 72.0 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 22.4 GB | 23.8 GB | 256K | 10 | Reference | Long context |
| Q8_0 | 11.9 GB | 13.3 GB | 256K | 19 | −0.1% ppl | Long context |
| Q6_K | 9.2 GB | 10.6 GB | 256K | 24 | −0.4% ppl | Long context |
| Q5_K_M | 7.9 GB | 9.3 GB | 256K | 28 | −0.8% ppl | Long context |
| Q4_K_M | 6.7 GB | 8.2 GB | 256K | 33 | −1.9% ppl | Recommended |
| Q3_K_M | 5.5 GB | 6.9 GB | 256K | 41 | −5.4% ppl | Long context |
| Q2_K | 4.7 GB | 6.1 GB | 256K | 48 | −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, 8 of 48 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-4-12B-it-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.
01Download is 6.7 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 72.0 GB of 96 GB. Raising it with iogpu.wired_limit_mb is possible, and risky.
03There is room to go to the model's full 256K context on this card.