Model comparison
14B vs Qwen3-Coder-30B-A3B-Instruct
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Qwen3-Coder-30B-A3B-Instruct's KV cache at 32K is 8.3× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| 14B | Qwen3-Coder-30B-A3B-Instruct | |
|---|---|---|
| Parameters | 14.2B | 30.5B |
| Architecture | llama | qwen3moe |
| Layers | 40 | 48 |
| Native context | 8,192 | 262,144 |
| Mixture of experts | no | yes, 128 experts |
| Quantizations published | 5 | 46 |
| Smallest quantization | 7.62 GiB | 7.46 GiB |
| Q4_K_M | — | 17.28 GiB |
| Licence | wtfpl | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | 14B | Qwen3-Coder-30B-A3B-Instruct | Ratio |
|---|---|---|---|
| 4,096 | 3.13 GiB | 0.38 GiB | 8.33× |
| 8,192 | 6.25 GiB | 0.75 GiB | 8.33× |
| 16,384 | 12.50 GiB | 1.50 GiB | 8.33× |
| 32,768 | 25.00 GiB | 3.00 GiB | 8.33× |
| 65,536 | 50.00 GiB | 6.00 GiB | 8.33× |
| 131,072 | 100.00 GiB | 12.00 GiB | 8.33× |