kimi-k2-0905-preview

Kimi K2-0905 is a high-performance hybrid expert large model launched by Moonshot AI, positioned as a developer-friendly intelligent foundation with the core strengths of "enhanced real-world programming capabilities + faster API response + an ultra-long context of 256K tokens."

• Outstanding Performance in Engineering Tasks: Significantly improves in real-world software engineering evaluations such as SWE-bench Verified, accurately understanding GitHub Issues and generating complete, runnable fix code.
• Enhanced Frontend Development Experience: The generated web code combines functionality with modern UI design, supports responsive layouts and mainstream frameworks, and can be directly used for product prototypes.
• Context Expansion Up to 256K Tokens: Supports full-project-level code analysis, processing of ultra-long documents, and complex multi-turn conversations, meeting the demands of large-scale engineering tasks.
• High-Speed API Support: The concurrently released kimi-k2-turbo-preview delivers an output speed of 60–100 tokens per second, balancing efficiency with powerful capabilities.
• Deep Compatibility with Open Platforms: Fully compatible with the Anthropic API, supporting WebSearch Tool, Token Enforcer, and automatic Context Caching, thereby reducing integration costs.

───────────────────────────────────────────────────────────────────

Core Capabilities

👨‍💻 Real-World Programming: Completes the entire PR submission process—from natural language descriptions to fully tested pull requests.

🎨 Production-Ready Frontend Generation: Outputs beautiful, functional, and modern web applications based on given instructions.

📚 Ultra-Long Context Understanding: Loads and analyzes codebases of up to 100,000 lines, providing support for refactoring or question-and-answer tasks.

⚡ High-Speed and Stable Inference: Offers a low-latency, high-throughput backend engine for AI programming tools.