Model comparison
MiniMax-H3 vs Wan2.2-Distill-Models
At Q4_K_M, Wan2.2-Distill-Models is the smaller download — 9,661,569,664 bytes against 14,576,977,960.
From the file· summed bytes, KV per layer
Side by side
| MiniMax-H3 | Wan2.2-Distill-Models | |
|---|---|---|
| Parameters | 33.1B | 14.3B |
| Architecture | qwen3vl | wan |
| Layers | — | — |
| Native context | — | — |
| Mixture of experts | no | no |
| Quantizations published | 22 | 67 |
| Smallest quantization | 10.60 GiB | 4.95 GiB |
| Q4_K_M | 13.58 GiB | 9.00 GiB |
| Licence | other | — |
KV cache by context
the term that decides long-context viability
| Context | MiniMax-H3 | Wan2.2-Distill-Models | Ratio |
|---|---|---|---|
| 4,096 | — | — | — |
| 8,192 | — | — | — |
| 16,384 | — | — | — |
| 32,768 | — | — | — |
| 65,536 | — | — | — |
| 131,072 | — | — | — |