Model comparison
Mixtral_34Bx2_MoE_60B vs Qwen3.6-27B
At Q4_K_M, Qwen3.6-27B is the smaller download — 16,547,397,888 bytes against 34,390,405,728. At long context the gap widens: Qwen3.6-27B's KV cache at 32K is 3.7× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Mixtral_34Bx2_MoE_60B | Qwen3.6-27B | |
|---|---|---|
| Parameters | 60.8B | 27.8B |
| Architecture | llama | qwen35 |
| Layers | 60 | 64 |
| Native context | 200,000 | 262,144 |
| Mixture of experts | yes, 2 experts | no |
| Quantizations published | 7 | 40 |
| Smallest quantization | 19.14 GiB | 8.74 GiB |
| Q4_K_M | 32.03 GiB | 15.41 GiB |
| Licence | cc-by-nc-4.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Mixtral_34Bx2_MoE_60B | Qwen3.6-27B | Ratio |
|---|---|---|---|
| 4,096 | 0.94 GiB | 0.25 GiB | 3.75× |
| 8,192 | 1.88 GiB | 0.50 GiB | 3.75× |
| 16,384 | 3.75 GiB | 1.00 GiB | 3.75× |
| 32,768 | 7.50 GiB | 2.00 GiB | 3.75× |
| 65,536 | 15.00 GiB | 4.00 GiB | 3.75× |
| 131,072 | 30.00 GiB | 8.00 GiB | 3.75× |