Model comparison
gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking vs Qwen3.6-35B-A3B
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Qwen3.6-35B-A3B's KV cache at 32K is 9.9× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking | Qwen3.6-35B-A3B | |
|---|---|---|
| Parameters | 31.3B | 36.0B |
| Architecture | gemma4 | qwen35moe |
| Layers | 60 | 40 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | no | yes, 256 experts |
| Quantizations published | 27 | 61 |
| Smallest quantization | 6.66 GiB | 8.77 GiB |
| Q4_K_M | — | 19.71 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking | Qwen3.6-35B-A3B | Ratio |
|---|---|---|---|
| 4,096 | 1.80 GiB | 0.08 GiB | 22.98× |
| 8,192 | 2.42 GiB | 0.16 GiB | 15.49× |
| 16,384 | 3.67 GiB | 0.31 GiB | 11.75× |
| 32,768 | 6.17 GiB | 0.63 GiB | 9.87× |
| 65,536 | 11.17 GiB | 1.25 GiB | 8.94× |
| 131,072 | 21.17 GiB | 2.50 GiB | 8.47× |