Kimi K3
Moonshot AI's agentic flagship, built for long-horizon tool use and very long context, with open weights under a Modified MIT license. A leading open model for autonomous, multi-step work.
At a glance
Released
Sep 30, 2026
Provider
Moonshot AI
Context window
512,000 tokens
Max output
64,000 tokens
Modalities
Text + image → text
Knowledge cutoff
June 2026
API model ID
kimi-k3License
Modified MIT
Pricing
Standard
Input$2.50 / 1M
Output$14 / 1M
* Rates are from a third-party aggregator — verify before relying on them. A typical 1,500-token prompt with a 650-token answer costs ≈ $0.013 on the standard tier. Run your own prompt through the calculator →
What's new vs the previous model
Compared withKimi K2.6 (July 2026)
- Stronger agentic planning and longer context than K2.6
- Higher price and heavier serving requirements
Best use cases
Autonomous agents
Executes long chains of tool calls while tracking goals and state.
Long-document work
Reasons over very large codebases and document sets in one pass.
Open self-hosting
Can be deployed privately under its permissive license.
Strengths & limitations
Strengths
- Strong agentic behavior
- Very long context
- Open weights
Limitations
- Premium pricing for an open model
- Resource-intensive to self-host
How to test it
- No-code: open kimi.com and pick Kimi K3 from the model switcher — no API key needed.
- Playground: sign in to the Moonshot AI Platform and run prompts in the workbench, then copy the generated code.
- API: create a key in the Moonshot AI Platform and call the API (example below).
curl https://api.moonshot.ai/v1/chat/completions \
# OpenAI-compatible endpoint
-H "Authorization: Bearer $API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k3",
"messages": [{"role": "user", "content": "Explain tokenization in one paragraph."}]
}'