Model comparison
Qwen3-Coder-30B-A3B-Instruct vs Valkyrie-49B-v2.1
At Q4_K_M, Qwen3-Coder-30B-A3B-Instruct is the smaller download — 18,556,689,568 bytes against 30,215,578,688. At long context the gap widens: Qwen3-Coder-30B-A3B-Instruct's KV cache at 32K is 26.7× 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 | Valkyrie-49B-v2.1 | |
|---|---|---|
| Parameters | 30.5B | 49.9B |
| Architecture | qwen3moe | deci |
| Layers | 48 | 80 |
| Native context | 262,144 | 131,072 |
| Mixture of experts | yes, 128 experts | no |
| Quantizations published | 46 | 50 |
| Smallest quantization | 7.46 GiB | 10.27 GiB |
| Q4_K_M | 17.28 GiB | 28.14 GiB |
| Licence | apache-2.0 | — |
KV cache by context
the term that decides long-context viability
| Context | Qwen3-Coder-30B-A3B-Instruct | Valkyrie-49B-v2.1 | Ratio |
|---|---|---|---|
| 4,096 | 0.38 GiB | 10.00 GiB | 26.67× |
| 8,192 | 0.75 GiB | 20.00 GiB | 26.67× |
| 16,384 | 1.50 GiB | 40.00 GiB | 26.67× |
| 32,768 | 3.00 GiB | 80.00 GiB | 26.67× |
| 65,536 | 6.00 GiB | 160.00 GiB | 26.67× |
| 131,072 | 12.00 GiB | 320.00 GiB | 26.67× |