Model comparison
Qwen3.8-27B vs llm-jp-4-32b-a3b-thinking
At Q4_K_M, Qwen3.8-27B is the smaller download — 16,810,714,336 bytes against 21,400,590,016.
From the file· summed bytes, KV per layer
Side by side
| Qwen3.8-27B | llm-jp-4-32b-a3b-thinking | |
|---|---|---|
| Parameters | 27.8B | 32.1B |
| Architecture | qwen35 | qwen3moe |
| Layers | 64 | 32 |
| Native context | 262,144 | 65,536 |
| Mixture of experts | no | yes, 128 experts |
| Quantizations published | 22 | 11 |
| Smallest quantization | 8.39 GiB | 15.60 GiB |
| Q4_K_M | 15.66 GiB | 19.93 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3.8-27B | llm-jp-4-32b-a3b-thinking | Ratio |
|---|---|---|---|
| 4,096 | 0.25 GiB | 0.25 GiB | 1.00× |
| 8,192 | 0.50 GiB | 0.50 GiB | 1.00× |
| 16,384 | 1.00 GiB | 1.00 GiB | 1.00× |
| 32,768 | 2.00 GiB | 2.00 GiB | 1.00× |
| 65,536 | 4.00 GiB | 4.00 GiB | 1.00× |
| 131,072 | 8.00 GiB | 8.00 GiB | 1.00× |