Kimi API Pricing (October 2026)
Kimi K3 costs $3 per million cache-miss input tokens and $15 per million output tokens, with a $0.30 cache-hit input rate and a 1M-token context window. The Kimi API costs $0.95/$4.00 for Kimi K2.6 and the coding-tuned Kimi K2.7 Code. Moonshot's pricing page no longer lists Kimi K2.5 or K2.
Moonshot's official Kimi K3 pricing page lists $3 cache-miss input, $0.30 cache-hit input, and $15 output per million tokens. Kimi K3 also raises the context ceiling to 1M. The previous flagship tier remains at $0.95/$4.00 (K2.6 and its coding sibling K2.7 Code) with 256K context. K2.5 and K2 are no longer on Moonshot's pricing page; the table keeps their last published rates, marked Previous. The table below owns the current Kimi price comparison; each rate is tied to Moonshot's platform documentation rather than a third-party router.
Start with the Kimi K3 model profile for current benchmark coverage, then compare the family in the Chinese model rankings, the price-vs-performance view, or the head-to-head model comparison.
Every Kimi API price per 1M tokens
| Model | Input $/M | Cached input $/M | Output $/M | Context |
|---|---|---|---|---|
| Kimi K3 | $3 | $0.3 | $15 | 1.05M |
| Kimi 2.6 | $0.95 | $0.16 | $4 | 256K |
| Kimi K2.7 Code | $0.95 | $0.19 | $4 | 256K |
| Kimi K2.5 | $0.6 | — | $3 | 256K |
| Kimi K2 | $0.6 | — | $2.5 | 128K |
The calculator uses real-time rates. Moonshot's Batch API charges 60% of standard price for K2.7 Code and K2.6, but the official Batch pricing page does not list Kimi K3. Cache-hit input is $0.30/M for Kimi K3, $0.19/M for K2.7 Code, and $0.16/M for K2.6.
Estimate your monthly Kimi API bill
| Model | Est. monthly cost |
|---|---|
| Kimi K3 | $60 |
| Kimi 2.6 | $17.5 |
| Kimi K2.7 Code | $17.5 |
Estimates use Kimi's standard per-token rates from the table above. For token-level presets and cross-provider comparison, use the full LLM API pricing calculator.
How Moonshot prices the Kimi API
Kimi K3 is the new premium API tier at $3 input / $15 output per million tokens, with cache-hit input at $0.30. The previous flagship rate of $0.95/$4.00 still covers both K2.6 (the general model) and kimi-k2.7-code (the coding endpoint), so choosing the specialist costs nothing extra. K2.5, which ran thinking and non-thinking modes under one model at one price, is no longer on Moonshot's pricing page.
The discount Moonshot publishes is cache-hit input: $0.30 per million tokens on Kimi K3, $0.19 on K2.7 Code, and $0.16 on K2.6. For coding agents that resend large repository context on every turn, that cache lane is where most of the real savings live. Moonshot's Batch pricing page also charges 60% of standard price for K2.7 Code and K2.6. Kimi K3 is not on the supported Batch model list, so the calculator above stays on comparable real-time rates instead of applying the discount to every row.
Kimi vs DeepSeek and OpenAI pricing
Against its closest rival, the price ordering now depends on the hour: at peak ($1.32/$3.96), DeepSeek V4 Pro costs about 39% more than K2.6 ($0.95/$4.00) on input and about 1% less on output. Off-peak ($0.66/$1.98) it is cheaper on both. That does not settle quality, latency, or deployment fit. For pure volume work, DeepSeek V4.1 Flash ($0.30/$1.20 at peak) costs less than every Kimi model Moonshot lists, on both rates.
Kimi K3's $15 output rate costs about 25% more than GPT-5.6 Terra's $12 and about 25% less than GPT-5.6 Sol's $20. K2.6 remains the cheaper Kimi endpoint at $4 output. See the full DeepSeek API pricing and OpenAI API pricing tables, or line models up head-to-head on the compare hub.
Which Kimi model to use
Kimi K3 is the capability pick and the only Kimi row with a 1M-token context window. Its model profile carries the current overall, coding, and agentic evidence rather than freezing those scores on a pricing page. If your workload is coding agents specifically, K2.7 Code is the same $0.95/$4.00 as K2.6 with a published $0.19 cache-hit input rate; its profile keeps specialist benchmark coverage separate from overall ranking eligibility.
Moonshot no longer lists K2.5 or the legacy Kimi K2, so K2.6 and K2.7 Code ($0.95/$4.00) now carry the lowest Kimi rate. Kimi K3's full weights are now published, but a 2.8T-parameter model is not a casual self-host. Read the Kimi K3 release review and run the self-host calculator before treating zero API spend as zero infrastructure cost.
Kimi API pricing FAQ
How much does the Kimi API cost?
Kimi K3 costs $3 input and $15 output per million tokens, with cache-hit input at $0.30. Kimi K2.6 and Kimi K2.7 Code both cost $0.95/$4.00. Batch jobs cost 60% of the standard rate for K2.7 Code and K2.6; Kimi K3 is not currently listed for Batch, and Moonshot no longer lists K2.5.
Is the Kimi K2 API free?
No — Moonshot's hosted Kimi API is pay-per-token. The genuinely free path is self-hosting: K2.6 is an open-weight release, so teams with their own GPUs can run it without per-token fees. Hardware, operations, and power still have a cost. Use the self-host calculator and open-source model rankings before choosing that route.
What is Moonshot AI's API pricing per million tokens?
Kimi K3 is $3 input and $15 output per million tokens. K2.6 and K2.7 Code cost $0.95/$4.00. Moonshot lists cache-hit input at $0.30 for Kimi K3, $0.19 for K2.7 Code, and $0.16 for K2.6. Eligible Batch jobs on K2.7 Code and K2.6 cost 60% of the corresponding real-time rate.
How does Kimi pricing compare to DeepSeek?
DeepSeek V4 Pro ($1.32/$3.96 at peak) costs about 39% more than Kimi K2.6 ($0.95/$4.00) on input and roughly the same on output. Off-peak ($0.66/$1.98), it undercuts K2.6 on both. Token price therefore depends on when the work runs, while model quality, latency, support, and deployment requirements can still favor Kimi. For high-volume pipelines DeepSeek V4.1 Flash ($0.30/$1.20 at peak) is cheaper again — see the DeepSeek pricing hub for the full table.