Model comparison
Qwen3-Coder-30B-A3B-Instruct vs Marco-Mini-Instruct
At Q4_K_M, Marco-Mini-Instruct is the smaller download — 10,505,424,448 bytes against 18,556,689,568. At long context the gap widens: Qwen3-Coder-30B-A3B-Instruct's KV cache at 32K is 1.2× 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 | Marco-Mini-Instruct | |
|---|---|---|
| Parameters | 30.5B | 17.3B |
| Architecture | qwen3moe | qwen3moe |
| Layers | 48 | 28 |
| Native context | 262,144 | 32,768 |
| Mixture of experts | yes, 128 experts | yes, 256 experts |
| Quantizations published | 46 | 35 |
| Smallest quantization | 7.46 GiB | 3.33 GiB |
| Q4_K_M | 17.28 GiB | 9.78 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 | Marco-Mini-Instruct | Ratio |
|---|---|---|---|
| 4,096 | 0.38 GiB | 0.44 GiB | 1.17× |
| 8,192 | 0.75 GiB | 0.88 GiB | 1.17× |
| 16,384 | 1.50 GiB | 1.75 GiB | 1.17× |
| 32,768 | 3.00 GiB | 3.50 GiB | 1.17× |
| 65,536 | 6.00 GiB | 7.00 GiB | 1.17× |
| 131,072 | 12.00 GiB | 14.00 GiB | 1.17× |