Model comparison
Trinity-Nano-Preview vs Qwen3-8B
At Q4_K_M, Trinity-Nano-Preview is the smaller download — 3,786,957,088 bytes against 5,027,783,488. At long context the gap widens: Trinity-Nano-Preview's KV cache at 32K is 8.3× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Trinity-Nano-Preview | Qwen3-8B | |
|---|---|---|
| Parameters | 6.1B | 8.2B |
| Architecture | afmoe | qwen3 |
| Layers | 56 | 36 |
| Native context | 131,072 | 40,960 |
| Mixture of experts | yes, 128 experts | no |
| Quantizations published | 50 | 51 |
| Smallest quantization | 1.94 GiB | 2.12 GiB |
| Q4_K_M | 3.53 GiB | 4.68 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Trinity-Nano-Preview | Qwen3-8B | Ratio |
|---|---|---|---|
| 4,096 | 0.16 GiB | 0.56 GiB | 3.58× |
| 8,192 | 0.21 GiB | 1.13 GiB | 5.31× |
| 16,384 | 0.32 GiB | 2.25 GiB | 7.00× |
| 32,768 | 0.54 GiB | 4.50 GiB | 8.33× |
| 65,536 | 0.98 GiB | 9.00 GiB | 9.21× |
| 131,072 | 1.85 GiB | 18.00 GiB | 9.72× |