KAT-Coder-Pro-V1

The Kuaishou KAT-Coder series is a large model family of Agentic Coding launched by Kuaishou's Kwaipilot team, with its core positioning as a "cost-effective AI programming assistant covering all scenarios". It includes three differentiated models, suitable for diverse needs ranging from personal lightweight development to enterprise-level complex projects, and the Air version is permanently free to lower the usage threshold.

• Differentiated layout of the model family: Covers three core models, KAT-Coder-Pro V1 (high-performance closed-source version, SWE-Bench Verified solution rate of 73.4%, 256K context, suitable for enterprise-level complex tasks), KAT-Coder-Air V1 (lightweight free version, ranked fifth globally among open-source models of the same scale, 128K context, ultra-fast response), KAT-Coder-Exp-72B (open-source experimental version, SWE-Bench Verified solution rate of 74.6%, topping the open-source SOTA, for academic exploration).
• Top-notch practical performance: The code generated by the Pro version can be directly run, such as a web-based operating system (including functions like text editor, drawing board, etc.), a Cyberpunk-style Fruit Ninja HTML5 game (completion rate exceeding Grok4 and Claude Sonnet 4.5); in the Kilo Code evaluation, it completed 10 functions in 4 minutes with zero tool call failures, and the total cost was only $0.072, with better cost-effectiveness than Haiku 4.5 and GLM 4.6.
• Wide tool and deployment compatibility: The entire series is compatible with mainstream AI programming tools such as Claude Code, Cline, Kilo Code, Roo Code, etc.; the Pro version is used via the Wanqing API, the Air version supports API and Hugging Face deployment (requiring 1×A100 or 4×RTX4090), the Exp-72B requires 2×A100 or 8×RTX4090, and supports over 20 mainstream programming languages.

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

Core capabilities

💻 Whole-link code generation: Covers the entire process from requirement analysis, architecture planning to coding implementation, debugging and error correction. The Pro version supports project-level development, and the Air version meets basic function writing.

🔧 Stable tool invocation: Low API call error rate, can call Git, compilers, debugging tools, etc., suitable for complex development task flows, with zero failures in the Kilo Code evaluation.

📚 Ultra-long context processing: The Pro version has 256K context, and the Air/Exp versions have 128K context, easily handling large-scale codebase understanding and long-text programming tasks.

🌍 Multi-scenario adaptation: Covers scenarios such as front-end (HTML/CSS/JS), back-end, algorithm implementation, etc., supports multiple languages, and the Pro version is suitable for industrial and enterprise-level engineering requirements.

⚡ High efficiency and low consumption: The Air version has ultra-fast response, and the Pro version has outstanding cost-effectiveness, consuming less than models of the same performance for the same task, reducing development costs.