Kimi API pricing, every Moonshot AI model per million tokens
On 10 October 2026, Moonshot AI models on this list cost from $0.49 per million input tokens (Kimi K2.5) to $3.00 (Kimi K3), and from $2.30 per million output tokens (Kimi K2 0711) to $15.00 (Kimi K3).
| Model | Input | Output | Cached input | Batch in / out | Context | Your month | Listed |
|---|---|---|---|---|---|---|---|
| Kimi K3 | $3.00 | $15.00 | $0.30 | $2.28 / $11.40 | 1.05M | $405.00 | 2026-07-16 |
| Kimi K2.7 Code | $0.95 | $4.00 | $0.19 | 262K | $117.00 | 2026-06-12 | |
| Kimi K2.6 | $0.95 | $4.00 | $0.16 | 262K | $117.00 | 2026-04-20 | |
| Kimi K2.5 AtlasCloud, cheapest of 5 hosts | $0.49 | $2.50 | $0.20 | 262K | $66.90 | 2026-01-27 | |
| Kimi K2 Thinking Google, cheapest of 2 hosts | $0.60 | $2.50 | 262K | $73.50 | 2025-11-06 | ||
| Kimi K2 0905 served by Novita | $0.60 | $2.50 | 262K | $73.50 | 2025-09-04 | ||
| Kimi K2 0711 served by Novita | $0.57 | $2.30 | 131K | $68.70 | 2025-07-11 |
4 of these 7 models are not served by Moonshot AI itself on OpenRouter, so their price is the cheapest regular price among the hosts that run them, and the host is named under the model. Compare Moonshot AI with every other provider
Common questions
How much does the Kimi API cost?
On 10 October 2026, Moonshot AI models on this list cost from $0.49 per million input tokens (Kimi K2.5) to $3.00 (Kimi K3), and from $2.30 per million output tokens (Kimi K2 0711) to $15.00 (Kimi K3).
What is the cheapest Kimi model?
For a typical mix of three input tokens to every output token, the cheapest is Kimi K2.5 at $0.49 per million input tokens and $2.50 per million output tokens.
What is the newest Kimi model on the list?
Kimi K3, listed on 16 July 2026, at $3.00 per million input tokens and $15.00 per million output tokens.
Which Kimi model has the largest context window?
Kimi K3, with 1.05M tokens of context.
Does Moonshot AI offer batch pricing?
Yes, 1 of its models on this list show a batch price, with batch input at a median of 76 percent of the standard price. Batch suits work that can wait for its answer.
Choosing a model for a production agent
Price is one input. Quality on your own cases, latency and the cost of retries decide the rest. Tell me what you are building.