Model comparison
gemma-4-26B-A4B-it vs Mistral-Small-3.2-24B-Instruct-2506-ultra-uncensored-heretic
At Q4_K_M, Mistral-Small-3.2-24B-Instruct-2506-ultra-uncensored-heretic is the smaller download — 14,333,922,464 bytes against 16,796,017,248. At long context the gap widens: gemma-4-26B-A4B-it's KV cache at 32K is 3.2× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| gemma-4-26B-A4B-it | Mistral-Small-3.2-24B-Instruct-2506-ultra-uncensored-heretic | |
|---|---|---|
| Parameters | 26.5B | 24.0B |
| Architecture | gemma4 | mistral3 |
| Layers | 30 | 40 |
| Native context | 262,144 | 131,072 |
| Mixture of experts | yes, 128 experts | no |
| Quantizations published | 46 | 40 |
| Smallest quantization | 8.99 GiB | 4.91 GiB |
| Q4_K_M | 15.64 GiB | 13.35 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | gemma-4-26B-A4B-it | Mistral-Small-3.2-24B-Instruct-2506-ultra-uncensored-heretic | Ratio |
|---|---|---|---|
| 4,096 | 0.45 GiB | 0.63 GiB | 1.39× |
| 8,192 | 0.61 GiB | 1.25 GiB | 2.06× |
| 16,384 | 0.92 GiB | 2.50 GiB | 2.72× |
| 32,768 | 1.54 GiB | 5.00 GiB | 3.24× |
| 65,536 | 2.79 GiB | 10.00 GiB | 3.58× |
| 131,072 | 5.29 GiB | 20.00 GiB | 3.78× |