Kimi K3
Kimi K3 is Moonshot AI’s flagship model, with 2.8 trillion parameters, native visual understanding and a 1,048,576-token context window.
Last updated
kimi-k3Key facts
At a glance
- Context window
- 1,048,576
- Parameters
- 2.8T
- Released
- July 16, 2026
- Output / 1M
- $15.00
Pricing
How much does Kimi K3 cost?
Kimi K3 costs $3.00 per million input tokens on a cache miss, $0.30 per million on a cache hit, and $15.00 per million output tokens. Prices exclude tax.
| Token type | Price per 1M tokens |
|---|---|
| Input — cache hit | $0.30 |
| Input — cache miss | $3.00 |
| Output | $15.00 |
What a real request costs
A 100,000-token prompt with a 5,000-token response costs $0.375 cold. Repeat that same prompt prefix so 80% of the input hits the cache and it drops to $0.159 — a 58% saving.
This is why headline per-token prices mislead on agent workloads, where the system prompt and tool definitions repeat on every call.
Capabilities
What can Kimi K3 do?
- Text inputsupported
- Image inputsupported
- Video inputnot supported
- Thinking modesupported
- Tool callingsupported
- JSON modesupported
- Structured outputsupported
- Partial modesupported
- Context cachingsupported
- Web searchsupported
- Open weightssupported
Notes
What the spec sheet leaves out
- Always reasons. Reasoning depth is set with the top-level
reasoning_effortfield:low,highormax. The default ismax. - Adds two API capabilities over earlier models: tool choice constraints (
tool_choice) and dynamically loaded tools. - Moonshot notes that
web_searchis mid-update and does not currently recommend relying on it. - Open weights were published in July 2026.
FAQ
Kimi K3 questions
What is Kimi K3?
Kimi K3 is Moonshot AI’s flagship model, with 2.8 trillion parameters, native visual understanding and a 1,048,576-token context window.
What is the Kimi K3 context window?
Kimi K3 has a 1,048,576-token context window. That total covers input and output combined, so a long prompt reduces the room left for the response.
How much does Kimi K3 cost?
Kimi K3 costs $3.00 per million input tokens on a cache miss, $0.30 per million on a cache hit, and $15.00 per million output tokens. Prices exclude tax.
Is Kimi K3 open source?
Moonshot AI published downloadable open weights for Kimi K3, so you can self-host it. That is not the same as an OSI-approved open-source licence — check the licence attached to the weight release before deploying commercially.
Sources
Where these figures come from
Every specification and price on this page is transcribed from Moonshot AI’s own documentation and re-checked on the date shown. Nothing is estimated.