Model comparison
Qwen3-Coder-30B-A3B-Instruct vs KAT-Coder-V2.5-Dev
At Q4_K_M, Qwen3-Coder-30B-A3B-Instruct is the smaller download — 18,556,689,568 bytes against 21,391,448,480. At long context the gap widens: KAT-Coder-V2.5-Dev's KV cache at 32K is 4.8× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen3-Coder-30B-A3B-Instruct | KAT-Coder-V2.5-Dev | |
|---|---|---|
| Parameters | 30.5B | 34.7B |
| Architecture | qwen3moe | qwen35moe |
| Layers | 48 | 40 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | yes, 128 experts | yes, 256 experts |
| Quantizations published | 46 | 29 |
| Smallest quantization | 7.46 GiB | 9.11 GiB |
| Q4_K_M | 17.28 GiB | 19.92 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3-Coder-30B-A3B-Instruct | KAT-Coder-V2.5-Dev | Ratio |
|---|---|---|---|
| 4,096 | 0.38 GiB | 0.08 GiB | 4.80× |
| 8,192 | 0.75 GiB | 0.16 GiB | 4.80× |
| 16,384 | 1.50 GiB | 0.31 GiB | 4.80× |
| 32,768 | 3.00 GiB | 0.63 GiB | 4.80× |
| 65,536 | 6.00 GiB | 1.25 GiB | 4.80× |
| 131,072 | 12.00 GiB | 2.50 GiB | 4.80× |