Side by side
Green marks the better value in each row. Read November 18, 2026.
| Kimi K3.1 Pro | Hermes 3 405B | |
|---|---|---|
| Refusals | ||
| How refusals were removed | Abliterated | Fine-tuned |
| Size and context | ||
| Parameterstotal | 2,780B | 405B |
| Active parametersper token | All | All |
| Context windowtokens | 1,048,576 | 131,072 |
| Max outputtokens | 943,718 | Not published |
| Inputs | ||
| Text | ||
| Images | ||
| Video | ||
| Where it runs | ||
| Hosts | Silk Compute | Unfil.AI |
| API formatsat the host | 3 of 3 | 1 of 3 |
| Runs locally | ||
Which to pick
Kimi K3.1 Pro
Pick Kimi K3.1 Pro for long context, images and video, and coding agents that speak Messages or Responses.
Hermes 3 405B
Pick Hermes 3 405B if you want a fine-tuned model rather than an abliterated one.
Against the other models
Size
Total parameters, billions. Log scale. Larger is more capable
Context window
Tokens the model holds at once. Higher is better
Price
Blended USD per 1M tokens, lowest listed host. Lower is better
AbliteratedFine-tunedHatched: weights only, no API listed