Moonshot
API Kimi
Moonshot's Kimi models — a frontier tier and a coding-specialised one.
2
modèles dans la gamme
Kimi splits by specialisation rather than by size: K3 is Moonshot's frontier model for long-context reasoning and agentic tool use, and K2.7 Code is tuned for repository-scale edits and agentic software tasks. The two are priced within about six percent of each other on input, so the choice is about what the model was trained for, not about budget.
En bref
- À partir de
- $1.48 / 1 M de jetons
Modèles de cette famille
Tarifs
| Modèle | Option | Prix |
|---|---|---|
| Kimi K3 | input | $1.48/ 1 M de jetons |
| Kimi K3 | output | $7.4/ 1 M de jetons |
| Kimi K3 | cacheRead | $0.148/ 1 M de jetons |
| Kimi K2.7 Code | input | $1.57/ 1 M de jetons |
| Kimi K2.7 Code | output | $7.13/ 1 M de jetons |
| Kimi K2.7 Code | cacheRead | $0.265/ 1 M de jetons |
Lequel choisir
Kimi K3
Long-context reasoning and agentic tool use — the general frontier tier.
Kimi K2.7 Code
Repository-scale code edits and agentic software work, where a coding-tuned model beats a general one.
Points forts
- A coding-specialised tier that is not just the general model with a different prompt.
- Input prices sit within about six percent of each other, so specialisation does not cost extra.
- Prompt caching on both tiers.
- OpenAI-protocol compatible.
Limites
- K3's cache reads are discounted to a tenth of input while K2.7 Code only reaches about a sixth, so caching pays back differently per tier.
- Output costs four and a half to five times input on both tiers.
- Only two tiers, both at a similar price point — no cheap volume option.
La description de cette gamme par l'éditeur : Moonshot AI — docs