Model comparison
gemma-4-26B-A4B-it vs Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING-X8b
At Q4_K_M, Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING-X8b is the smaller download — 5,629,108,832 bytes against 16,796,017,248. At long context the gap widens: Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING-X8b's KV cache at 32K is 1.5× 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 | Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING-X8b | |
|---|---|---|
| Parameters | 26.5B | 9.4B |
| Architecture | gemma4 | qwen35 |
| Layers | 30 | 32 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | yes, 128 experts | no |
| Quantizations published | 46 | 5 |
| Smallest quantization | 8.99 GiB | 4.19 GiB |
| Q4_K_M | 15.64 GiB | 5.24 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 | Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING-X8b | Ratio |
|---|---|---|---|
| 4,096 | 0.45 GiB | 0.13 GiB | 3.59× |
| 8,192 | 0.61 GiB | 0.25 GiB | 2.42× |
| 16,384 | 0.92 GiB | 0.50 GiB | 1.84× |
| 32,768 | 1.54 GiB | 1.00 GiB | 1.54× |
| 65,536 | 2.79 GiB | 2.00 GiB | 1.40× |
| 131,072 | 5.29 GiB | 4.00 GiB | 1.32× |