Model comparison
Huihui-Qwen3-Coder-Next-abliterated vs Qwen3-30B-A3B-Thinking-2507
At Q4_K_M, Qwen3-30B-A3B-Thinking-2507 is the smaller download — 18,556,685,824 bytes against 48,556,632,000. At long context the gap widens: Huihui-Qwen3-Coder-Next-abliterated's KV cache at 32K is 4.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Huihui-Qwen3-Coder-Next-abliterated | Qwen3-30B-A3B-Thinking-2507 | |
|---|---|---|
| Parameters | 79.7B | 30.5B |
| Architecture | qwen3next | qwen3moe |
| Layers | 48 | 48 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | yes, 512 experts | yes, 128 experts |
| Quantizations published | 73 | 51 |
| Smallest quantization | 15.07 GiB | 7.05 GiB |
| Q4_K_M | 45.22 GiB | 17.28 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Huihui-Qwen3-Coder-Next-abliterated | Qwen3-30B-A3B-Thinking-2507 | Ratio |
|---|---|---|---|
| 4,096 | 0.09 GiB | 0.38 GiB | 4.00× |
| 8,192 | 0.19 GiB | 0.75 GiB | 4.00× |
| 16,384 | 0.38 GiB | 1.50 GiB | 4.00× |
| 32,768 | 0.75 GiB | 3.00 GiB | 4.00× |
| 65,536 | 1.50 GiB | 6.00 GiB | 4.00× |
| 131,072 | 3.00 GiB | 12.00 GiB | 4.00× |