Model comparison
Gemma-4-31B-Isometry-Fabled-Persona vs gpt-oss-20b
At Q4_K_M, gpt-oss-20b is the smaller download — 11,624,759,488 bytes against 18,687,065,824. At long context the gap widens: gpt-oss-20b's KV cache at 32K is 8.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Gemma-4-31B-Isometry-Fabled-Persona | gpt-oss-20b | |
|---|---|---|
| Parameters | 31.3B | 21.5B |
| Architecture | gemma4 | gpt-oss |
| Layers | 60 | 24 |
| Native context | 262,144 | 131,072 |
| Mixture of experts | no | yes, 32 experts |
| Quantizations published | 34 | 15 |
| Smallest quantization | 6.67 GiB | 10.68 GiB |
| Q4_K_M | 17.40 GiB | 10.83 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Gemma-4-31B-Isometry-Fabled-Persona | gpt-oss-20b | Ratio |
|---|---|---|---|
| 4,096 | 1.80 GiB | 0.11 GiB | 16.14× |
| 8,192 | 2.42 GiB | 0.21 GiB | 11.81× |
| 16,384 | 3.67 GiB | 0.39 GiB | 9.35× |
| 32,768 | 6.17 GiB | 0.77 GiB | 8.04× |
| 65,536 | 11.17 GiB | 1.52 GiB | 7.36× |
| 131,072 | 21.17 GiB | 3.02 GiB | 7.02× |