Model comparison
QwenPaw-Flash-9B vs Qwen3-30B-A3B-Thinking-2507
At Q4_K_M, QwenPaw-Flash-9B is the smaller download — 5,629,109,536 bytes against 18,556,685,824. At long context the gap widens: QwenPaw-Flash-9B's KV cache at 32K is 3.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| QwenPaw-Flash-9B | Qwen3-30B-A3B-Thinking-2507 | |
|---|---|---|
| Parameters | 9.4B | 30.5B |
| Architecture | qwen35 | qwen3moe |
| Layers | 32 | 48 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | no | yes, 128 experts |
| Quantizations published | 40 | 51 |
| Smallest quantization | 3.36 GiB | 7.05 GiB |
| Q4_K_M | 5.24 GiB | 17.28 GiB |
| Licence | — | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | QwenPaw-Flash-9B | Qwen3-30B-A3B-Thinking-2507 | Ratio |
|---|---|---|---|
| 4,096 | 0.13 GiB | 0.38 GiB | 3.00× |
| 8,192 | 0.25 GiB | 0.75 GiB | 3.00× |
| 16,384 | 0.50 GiB | 1.50 GiB | 3.00× |
| 32,768 | 1.00 GiB | 3.00 GiB | 3.00× |
| 65,536 | 2.00 GiB | 6.00 GiB | 3.00× |
| 131,072 | 4.00 GiB | 12.00 GiB | 3.00× |