Model comparison
Qwen3.5-9B-Claude-Opus-4.6-Distill vs gemma-4-12B-it-qat-q4_0-unquantized
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Qwen3.5-9B-Claude-Opus-4.6-Distill's KV cache at 32K is 2.5× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen3.5-9B-Claude-Opus-4.6-Distill | gemma-4-12B-it-qat-q4_0-unquantized | |
|---|---|---|
| Parameters | 9.0B | 12.0B |
| Architecture | qwen35 | gemma4 |
| Layers | 32 | 48 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 7 | 2 |
| Smallest quantization | 3.56 GiB | 6.50 GiB |
| Q4_K_M | 5.24 GiB | — |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3.5-9B-Claude-Opus-4.6-Distill | gemma-4-12B-it-qat-q4_0-unquantized | Ratio |
|---|---|---|---|
| 4,096 | 0.13 GiB | 0.72 GiB | 5.75× |
| 8,192 | 0.25 GiB | 0.97 GiB | 3.87× |
| 16,384 | 0.50 GiB | 1.47 GiB | 2.94× |
| 32,768 | 1.00 GiB | 2.47 GiB | 2.47× |
| 65,536 | 2.00 GiB | 4.47 GiB | 2.23× |
| 131,072 | 4.00 GiB | 8.47 GiB | 2.12× |