Model comparison
gemma-4-12B-it-qat-q4_0-unquantized vs Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16's KV cache at 32K is 3.9× 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 | Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16 | |
|---|---|---|
| Parameters | 12.0B | 35.1B |
| Architecture | gemma4 | qwen35moe |
| Layers | 48 | 40 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | no | yes, 256 experts |
| Quantizations published | 2 | 16 |
| Smallest quantization | 6.50 GiB | 7.67 GiB |
| Q4_K_M | — | 20.98 GiB |
| Licence | apache-2.0 | other |
KV cache by context
the term that decides long-context viability
| Context | gemma-4-12B-it-qat-q4_0-unquantized | Ornith-1.0-35B-AEON-Ultimate-Uncensored-BF16 | Ratio |
|---|---|---|---|
| 4,096 | 0.72 GiB | 0.08 GiB | 9.19× |
| 8,192 | 0.97 GiB | 0.16 GiB | 6.20× |
| 16,384 | 1.47 GiB | 0.31 GiB | 4.70× |
| 32,768 | 2.47 GiB | 0.63 GiB | 3.95× |
| 65,536 | 4.47 GiB | 1.25 GiB | 3.57× |
| 131,072 | 8.47 GiB | 2.50 GiB | 3.39× |