All AI models
Open-weightVerified October 23, 2026

Moonshot AI

Kimi K3

Kimi K3 is the right model to investigate when your work needs one million tokens of context, agentic coding, and native image understanding, and you can accept a frontier-scale deployment footprint or use a managed provider. Its custom license requires deliberate commercial review before productizing a hosted service.

What changed: Newer from Moonshot: Kimi K3.1 Pro is on the Kimi API at $2 input and $10 output per 1M tokens ($4 / $20 for K3.1 Pro Highspeed). As of October 23, 2026, Moonshot says the full K3.1 Pro weights will be released by November 4, 2026.
License
Kimi K3 License
Context
1,048,576 tokens
Modalities
Text · Image
Parameters
2.8T total, 104B active

What the license means

The full weights are available under Moonshot AI's custom Kimi K3 License. Model-as-a-service businesses above the stated revenue threshold and products above the stated scale threshold have additional obligations, so this is not a blanket Apache or MIT release.

Open-weight, open source, and no-cost API access mean different things. Always review the linked terms and your own data, hosting, and compliance requirements before shipping.

API and operating cost

Input / 1M
$3
Output / 1M
$15

kimi-k3 on the Kimi platform, per 1M tokens. Cached input costs $0.30. Kimi K3.1 Pro costs less at $2 / $10.

Check current provider price

Where it fits

  • Long-horizon coding agents
  • Large repository and document analysis
  • Multimodal knowledge work

Where it does not fit

  • Offline laptop experimentation
  • Teams that need a fully permissive license
  • Low-latency, small-context chat

Local deployment reality

Official documentation recommends vLLM, SGLang, or TokenSpeed. At 2.8T total parameters, this is infrastructure-scale deployment, not a realistic local laptop model.

vLLMSGLangTokenSpeed

Practical setup notes

  • Preserve the complete assistant message, including reasoning and tool-call fields, on multi-turn API requests.
  • Use lower reasoning effort only after evaluating quality and latency on your actual workload.

Capability notes

Reasoning
Always on, with low, high, and max effort settings
Tools and output
Tool calls, structured output, official tools, and preserved reasoning history

Limitations to plan around

  • The total model size makes self-hosting a serious infrastructure project.
  • Commercial use conditions differ from permissive open-source licenses.

Continue comparing

Related model decisions

See all reviewed models
Open-weight

MiniMax M3

Agentic coding

Open source

GLM-5.2

Long-horizon coding agents

Open-weight

DeepSeek V4

Cost-sensitive API workflows

Editorial note: This is a dated decision page, not a benchmark leaderboard or a substitute for legal, security, or infrastructure review. Model and pricing facts can change after October 23, 2026; use the linked primary sources before committing budget or customer data.Rechecked October 23, 2026: license, parameter count, context window, and API price. Runtime, reasoning, and tool notes date from the August 11, 2026 audit.