Model comparison
gemma-4-26B-A4B-it vs gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: gemma-4-26B-A4B-it's KV cache at 32K is 4.0× 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 | gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic | |
|---|---|---|
| Parameters | 26.5B | 31.3B |
| Architecture | gemma4 | gemma4 |
| Layers | 30 | 60 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | yes, 128 experts | no |
| Quantizations published | 46 | 2 |
| Smallest quantization | 8.99 GiB | 16.44 GiB |
| Q4_K_M | 15.64 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 | gemma-4-31B-it-qat-q4_0-unquantized-uncensored-heretic | Ratio |
|---|---|---|---|
| 4,096 | 0.45 GiB | 1.80 GiB | 4.00× |
| 8,192 | 0.61 GiB | 2.42 GiB | 4.00× |
| 16,384 | 0.92 GiB | 3.67 GiB | 4.00× |
| 32,768 | 1.54 GiB | 6.17 GiB | 4.00× |
| 65,536 | 2.79 GiB | 11.17 GiB | 4.00× |
| 131,072 | 5.29 GiB | 21.17 GiB | 4.00× |