Model comparison
Magistral-Small-2509-Vision vs Qwen3.6-35B-A3B
At Q4_K_M, Magistral-Small-2509-Vision is the smaller download — 12,621,946,560 bytes against 21,166,757,728. At long context the gap widens: Qwen3.6-35B-A3B's KV cache at 32K is 8.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Magistral-Small-2509-Vision | Qwen3.6-35B-A3B | |
|---|---|---|
| Parameters | 24.0B | 36.0B |
| Architecture | llama | qwen35moe |
| Layers | 40 | 40 |
| Native context | 131,072 | 262,144 |
| Mixture of experts | no | yes, 256 experts |
| Quantizations published | 16 | 61 |
| Smallest quantization | 8.09 GiB | 8.77 GiB |
| Q4_K_M | 11.76 GiB | 19.71 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Magistral-Small-2509-Vision | Qwen3.6-35B-A3B | Ratio |
|---|---|---|---|
| 4,096 | 0.63 GiB | 0.08 GiB | 7.99× |
| 8,192 | 1.25 GiB | 0.16 GiB | 8.00× |
| 16,384 | 2.50 GiB | 0.31 GiB | 8.00× |
| 32,768 | 5.00 GiB | 0.63 GiB | 8.00× |
| 65,536 | 10.00 GiB | 1.25 GiB | 8.00× |
| 131,072 | 20.00 GiB | 2.50 GiB | 8.00× |