Model comparison
gemma-4-12B-it-qat-q4_0-unquantized vs gpt-oss-20b-BF16
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: gpt-oss-20b-BF16's KV cache at 32K is 3.2× 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 | gpt-oss-20b-BF16 | |
|---|---|---|
| Parameters | 12.0B | 20.9B |
| Architecture | gemma4 | gpt-oss |
| Layers | 48 | 24 |
| Native context | 262,144 | 131,072 |
| Mixture of experts | no | yes, 32 experts |
| Quantizations published | 2 | 8 |
| Smallest quantization | 6.50 GiB | 12.03 GiB |
| Q4_K_M | — | 14.72 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 | gpt-oss-20b-BF16 | Ratio |
|---|---|---|---|
| 4,096 | 0.72 GiB | 0.11 GiB | 6.46× |
| 8,192 | 0.97 GiB | 0.21 GiB | 4.72× |
| 16,384 | 1.47 GiB | 0.39 GiB | 3.74× |
| 32,768 | 2.47 GiB | 0.77 GiB | 3.22× |
| 65,536 | 4.47 GiB | 1.52 GiB | 2.94× |
| 131,072 | 8.47 GiB | 3.02 GiB | 2.81× |