Model comparison
gemma-4-12B-it-qat-q4_0-unquantized vs GLM-4.7-Flash-Claude-Opus-4.5-High-Reasoning-Distill
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: GLM-4.7-Flash-Claude-Opus-4.5-High-Reasoning-Distill'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-4-12B-it-qat-q4_0-unquantized | GLM-4.7-Flash-Claude-Opus-4.5-High-Reasoning-Distill | |
|---|---|---|
| Parameters | 12.0B | 31.2B |
| Architecture | gemma4 | deepseek2 |
| Layers | 48 | 47 |
| Native context | 262,144 | 202,752 |
| Mixture of experts | no | yes, 64 experts |
| Quantizations published | 2 | 13 |
| Smallest quantization | 6.50 GiB | 9.22 GiB |
| Q4_K_M | — | 16.89 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | gemma-4-12B-it-qat-q4_0-unquantized | GLM-4.7-Flash-Claude-Opus-4.5-High-Reasoning-Distill | Ratio |
|---|---|---|---|
| 4,096 | 0.72 GiB | 0.21 GiB | 3.48× |
| 8,192 | 0.97 GiB | 0.41 GiB | 2.35× |
| 16,384 | 1.47 GiB | 0.83 GiB | 1.78× |
| 32,768 | 2.47 GiB | 1.65 GiB | 1.49× |
| 65,536 | 4.47 GiB | 3.30 GiB | 1.35× |
| 131,072 | 8.47 GiB | 6.61 GiB | 1.28× |