Provider pricing hub
Kimi API Pricing (July 2026)
The Kimi API costs $0.95/$4.00 per million input/output tokens for Kimi K2.6 and the coding-tuned Kimi K2.7 Code, and $0.60/$3.00 for Kimi K2.5. All three current-generation models ship a 256K-token context window, and Moonshot's published cache-hit input rate on K2.7 Code is $0.19 per million tokens.
Moonshot AI keeps Kimi pricing unusually simple: a flagship tier at $0.95/$4.00 (K2.6 and its coding sibling K2.7 Code) and a value tier at $0.60/$3.00 (K2.5), all with 256K context. There are no per-prompt-length surcharges and no separate thinking SKU — Moonshot documents K2.5 as one model that supports both thinking and non-thinking modes at the same price. The table below lists every Kimi model in BenchLM's registry, synced from Moonshot's official Kimi platform pricing pages.
Price is only half the story: K2.6 is currently the top open-weight model on BenchLM's coding leaderboard, and the Kimi rows feature prominently in our Chinese model rankings and the price-vs-performance view.
Every Kimi API price per 1M tokens
| Model | Input $/M | Cached input $/M | Output $/M | Context |
|---|---|---|---|---|
| Kimi 2.6 | $0.95 | $0.095 | $4 | 256K |
| Kimi K2.7 Code | $0.95 | $0.095 | $4 | 256K |
| Kimi K2.5 | $0.6 | $0.06 | $3 | 256K |
| Kimi K2 | $0.6 | $0.06 | $2.5 | 128K |
The cached-input column is an estimate at 10% of the fresh input rate; Moonshot's published cache-hit prices are $0.19/M on K2.7 Code and $0.15/M on Kimi K2. Moonshot does not publish a Batch API discount tier, and K2.5's thinking and non-thinking modes share one price and are collapsed into one row.
Estimate your monthly Kimi API bill
| Model | Est. monthly cost |
|---|---|
| Kimi 2.6 | $17.5 |
| Kimi K2.7 Code | $17.5 |
| Kimi K2.5 | $12 |
| Kimi K2 | $11 |
Estimates use Kimi's standard per-token rates from the table above. For task-level presets and cross-provider comparison, use the full AI cost calculator.
How Moonshot prices the Kimi API
Everything on Moonshot's price sheet is flat per-token billing. The flagship rate of $0.95 input / $4.00 output covers both K2.6 (the general model) and kimi-k2.7-code (the coding endpoint), so choosing the specialist costs nothing extra. K2.5 sits at $0.60/$3.00, and Moonshot explicitly documents that its thinking and non-thinking modes run under the same model at the same price — no separate reasoning surcharge to model.
The one discount Moonshot does publish is cache-hit input: $0.19 per million tokens on K2.7 Code (an 80% cut from the $0.95 cache-miss rate) and $0.15 on the older Kimi K2. For coding agents that resend large repository context on every turn, that cache lane is where most of the real savings live. There is no Batch API tier — if you need asynchronous half-price processing, Kimi doesn't offer it yet.
Kimi vs DeepSeek and OpenAI pricing
Against its closest rival, the picture is split: K2.6 at $0.95/$4.00 undercuts DeepSeek V4 Pro ($1.74/$3.48) by ~45% on input but costs slightly more on output, so output-heavy workloads tilt toward DeepSeek and prompt-heavy ones toward Kimi. For pure volume work, DeepSeek V4 Flash ($0.14/$0.28) undercuts every Kimi tier by a wide margin.
Against Western flagships the gap is one-directional: K2.6's $4.00 output is 3.75x cheaper than GPT-5.6 Terra ($15) and 7.5x cheaper than GPT-5.6 Sol ($30). 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 K2.6is the default: it scores 75 on BenchLM's provisional leaderboard and is the top open-weight model on our coding leaderboard at 88.1 — within six points of the overall coding leader. If your workload is coding agents specifically, K2.7 Codeis the same $0.95/$4.00 with a published $0.19 cache-hit input rate, though it doesn't yet have a ranked BenchLM score of its own.
K2.5 ($0.60/$3.00) is the value tier — 63 overall as the non-thinking row and 69 in reasoning mode, still with 256K context. The legacy Kimi K2 ($0.60/$2.50, 128K) remains on the price sheet mostly for existing integrations. And because the flagship weights are published, teams with GPUs can self-host instead of paying per token.
Kimi API pricing FAQ
How much does the Kimi API cost?
Kimi K2.6 and Kimi K2.7 Code both cost $0.95 input / $4.00 output per million tokens, and Kimi K2.5 costs $0.60/$3.00 — all with 256K context windows. The older Kimi K2 remains available at $0.60/$2.50 with 128K context, and Moonshot publishes cache-hit input rates of $0.19/M (K2.7 Code) and $0.15/M (K2).
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. It currently tops the open-weight rows on BenchLM's coding leaderboard; see the open-source model rankings for alternatives.
What is Moonshot AI's API pricing per million tokens?
Two tiers cover the current lineup: $0.95 input / $4.00 output for the flagship K2.6 and K2.7 Code, and $0.60/$3.00 for K2.5. There are no batch discounts and no thinking-mode surcharge — K2.5's thinking and non-thinking modes share one price. Cache-hit input, where published, cuts the input rate by 75–80%.
How does Kimi pricing compare to DeepSeek?
Kimi K2.6 ($0.95/$4.00) is ~45% cheaper than DeepSeek V4 Pro ($1.74/$3.48) on input but ~15% more expensive on output, so the winner depends on your input/output mix. For high-volume pipelines DeepSeek V4 Flash ($0.14/$0.28) is far cheaper than any Kimi tier — see the DeepSeek pricing hub for the full table.