Model comparison
Qwen3.8-27B vs Qwen3-Omni-30B-A3B-Instruct
At Q4_K_M, Qwen3.8-27B is the smaller download — 16,810,714,336 bytes against 18,557,053,952.
From the file· summed bytes, KV per layer
Side by side
| Qwen3.8-27B | Qwen3-Omni-30B-A3B-Instruct | |
|---|---|---|
| Parameters | 27.8B | 35.3B |
| Architecture | qwen35 | qwen3vlmoe |
| Layers | 64 | — |
| Native context | 262,144 | — |
| Mixture of experts | no | no |
| Quantizations published | 22 | 6 |
| Smallest quantization | 8.39 GiB | 17.28 GiB |
| Q4_K_M | 15.66 GiB | 17.28 GiB |
| Licence | apache-2.0 | — |
KV cache by context
the term that decides long-context viability
| Context | Qwen3.8-27B | Qwen3-Omni-30B-A3B-Instruct | Ratio |
|---|---|---|---|
| 4,096 | 0.25 GiB | — | — |
| 8,192 | 0.50 GiB | — | — |
| 16,384 | 1.00 GiB | — | — |
| 32,768 | 2.00 GiB | — | — |
| 65,536 | 4.00 GiB | — | — |
| 131,072 | 8.00 GiB | — | — |