Moonshot AI’s flagship multimodal model for long-horizon coding, deep reasoning, and end-to-end knowledge work.
2.8 trillion parameters
One of the largest open-weight model architectures developed to date
1M-token context
Processes extensive codebases, documents, conversations, and project histories
Long-horizon coding
Sustains complex engineering tasks with minimal human supervision
Native multimodal understanding
Analyzes text, images, screenshots, diagrams, and video inputs
Deep reasoning
Always uses reasoning mode for complex analytical and technical tasks
Agentic knowledge work
Completes multi-step research, analysis, and professional workflows
Large codebase understanding
Navigates, modifies, and reasons across repository-scale projects
Visual software engineering
Uses screenshots and visual feedback for frontend, game, and CAD workflows
Tool calling
Connects with custom functions and external tools during agent workflows
Dynamic tool loading
Makes new tools available during an ongoing conversation
Structured output
Supports strict JSON Schema responses for reliable application integration
Automatic context caching
Reuses unchanged prompt context without manual cache configuration
Large output capacity
Supports responses of up to approximately one million tokens
- Build and refactor applications across large codebases
- Run long-lived autonomous software engineering agents
- Debug complex issues and coordinate terminal-based development tools
- Create frontend interfaces using screenshots and visual feedback
- Develop games and interactive experiences from natural-language requirements
- Analyze large document collections and technical specifications
- Conduct deep research and produce structured professional reports
- Process images, diagrams, screenshots, and video content
- Build AI agents that call custom functions and external services
- Generate structured JSON outputs for production applications
- Automate multi-stage knowledge-work and business workflows
- Work with extensive project histories using a one-million-token context window
Available through the Kimi API Platform with usage-based pricing.
API pricing per 1 million tokens:
Cached input: $0.30
Input: $3.00
Output: $15.00
Pricing excludes applicable taxes and may vary by jurisdiction.
Kimi K3 is designed as an open-source model in the three-trillion-parameter class. The complete model weights are scheduled for release by July 27, 2026.
Until the weights are published, API and hosted Kimi services are the primary access methods.
Commercial use is subject to Moonshot AI’s applicable platform terms, model license, and usage policies.
Integration highlights
- Kimi API Platform
- Kimi.com
- Kimi Code CLI
- OpenAI SDK
- Claude Code
- Codex
- OpenCode
- OpenClaw
- Hermes Agent
- Custom Function Calling
- Official Kimi Tools