Model comparison
Qwen3.6-27B vs GigaChat3-10B-A1.8B-base
At Q4_K_M, GigaChat3-10B-A1.8B-base is the smaller download — 6,474,702,976 bytes against 16,547,397,888. At long context the gap widens: GigaChat3-10B-A1.8B-base's KV cache at 32K is 2.2× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen3.6-27B | GigaChat3-10B-A1.8B-base | |
|---|---|---|
| Parameters | 27.8B | 11.5B |
| Architecture | qwen35 | deepseek2 |
| Layers | 64 | 26 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | no | yes, 64 experts |
| Quantizations published | 40 | 8 |
| Smallest quantization | 8.74 GiB | 6.03 GiB |
| Q4_K_M | 15.41 GiB | 6.03 GiB |
| Licence | apache-2.0 | mit |
KV cache by context
the term that decides long-context viability
| Context | Qwen3.6-27B | GigaChat3-10B-A1.8B-base | Ratio |
|---|---|---|---|
| 4,096 | 0.25 GiB | 0.11 GiB | 2.19× |
| 8,192 | 0.50 GiB | 0.23 GiB | 2.19× |
| 16,384 | 1.00 GiB | 0.46 GiB | 2.19× |
| 32,768 | 2.00 GiB | 0.91 GiB | 2.19× |
| 65,536 | 4.00 GiB | 1.83 GiB | 2.19× |
| 131,072 | 8.00 GiB | 3.66 GiB | 2.19× |