Model comparison
pixtral-12b vs gemma-4-12B-it
At Q4_K_M, gemma-4-12B-it is the smaller download — 7,121,861,440 bytes against 7,477,204,608. At long context the gap widens: gemma-4-12B-it's KV cache at 32K is 2.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| pixtral-12b | gemma-4-12B-it | |
|---|---|---|
| Parameters | 12.7B | 12.0B |
| Architecture | llama | gemma4 |
| Layers | 40 | 48 |
| Native context | 1,024,000 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 32 | 59 |
| Smallest quantization | 3.85 GiB | 3.92 GiB |
| Q4_K_M | 6.96 GiB | 6.63 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | pixtral-12b | gemma-4-12B-it | Ratio |
|---|---|---|---|
| 4,096 | 0.63 GiB | 0.72 GiB | 1.15× |
| 8,192 | 1.25 GiB | 0.97 GiB | 1.29× |
| 16,384 | 2.50 GiB | 1.47 GiB | 1.70× |
| 32,768 | 5.00 GiB | 2.47 GiB | 2.03× |
| 65,536 | 10.00 GiB | 4.47 GiB | 2.24× |
| 131,072 | 20.00 GiB | 8.47 GiB | 2.36× |