Home
Kuaishou KwaiKAT Unveils KAT-Coder-Pro V2.5, First Domestic Agentic Coding Model for End-to-End Projects
Today, the KwaiKAT team officially launched KAT-Coder-Pro V2.5, its flagship Agentic Coding model. This new version introduces systematic upgrades across three key dimensions: long-term engineering capabilities, general Agentic capabilities, and a large-scale Agentic reinforcement learning system. The goal is to evolve from simple code completion to an AI agent that can independently handle complete software engineering tasks and complex business workflows.

For long-term engineering capabilities, the team built a custom AutoBuilder automated pipeline. This increased the success rate of creating runnable repository environments from an industry average of roughly 16.5% to 57.2%, covering 12 programming languages and over 100,000 verified real repository environments. Additionally, high-value failure trajectories are recycled into training data, enabling the model to master cross-file positioning, adherence to project specifications, and self-debugging testing.
For general Agentic capabilities, KwaiKAT developed the KwaiClawEnv system—a dynamically expanding tool pool. It derives numerous complex workflows from real business tasks and retains high-quality training trajectories through dual filtering. The system covers scenarios like data analysis, cross-system integration, and batch document processing, supporting task execution across more than 10 steps.

On the training side, the team moved away from pure supervised fine-tuning and adopted large-scale Agentic reinforcement learning. They used Harness Scaling to train across multiple mainstream Agent frameworks, preventing overfitting to a single interaction format. An asymmetric PPO architecture was introduced to address credit assignment in long-term tasks. A hierarchical reward mechanism was designed (core task results + behavioral norm constraints + failure exploration incentives) to balance effectiveness and robustness. The model also employs MOPD multi-teacher online strategy distillation to integrate capabilities from five expert models: long-term engineering, general Agentic, terminal usage, front-end aesthetics, and general knowledge. As a result, a single model can now handle multiple scenarios—writing code, running workflows, and generating front-end pages—without needing to switch.
Official evaluation data reveals: in code engineering, the SWE-Bench Pro score is 65.2, and the internal KAT Code Bench score is 53.1, enabling direct handling of complete Issues without manual decomposition. For Agentic tasks, the PinchBench score is 94.2, and the internal KAT Claw Bench score is 85.5, demonstrating excellent full-process stability.
KAT-Coder-Pro V2.5 is now fully available on the StreamLake platform (streamlake.com). It is open for API applications and technical documentation access, and the team has also released technical reports and developer exchange groups.
URL: https://streamlake.com/product/kat-coder
Related article
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation
Swedish AI Startup Lovable Eyes $13.2 Billion Valuation After Major Funding Round
As AI-driven coding tools gain traction, Swedish startup Lovable has secured a major funding round. The company aims to raise $3 billion, potentially boosting its valuation to $13.2 billion—double the $6.6 billion recorded last December. Menlo Ventur
Google Tests Remy AI Agent for Gemini as Focus Shifts to User Control
According to Business Insider, Google is testing Remy, a new AI personal agent for Gemini. This tool aims to execute tasks on behalf of users, streamlining both professional workflows and daily routines.Currently, Remy is undergoing testing in an int
Related Special Topic Recommendations
Comments (0)
0/500
Today, the KwaiKAT team officially launched KAT-Coder-Pro V2.5, its flagship Agentic Coding model. This new version introduces systematic upgrades across three key dimensions: long-term engineering capabilities, general Agentic capabilities, and a large-scale Agentic reinforcement learning system. The goal is to evolve from simple code completion to an AI agent that can independently handle complete software engineering tasks and complex business workflows.

For long-term engineering capabilities, the team built a custom AutoBuilder automated pipeline. This increased the success rate of creating runnable repository environments from an industry average of roughly 16.5% to 57.2%, covering 12 programming languages and over 100,000 verified real repository environments. Additionally, high-value failure trajectories are recycled into training data, enabling the model to master cross-file positioning, adherence to project specifications, and self-debugging testing.
For general Agentic capabilities, KwaiKAT developed the KwaiClawEnv system—a dynamically expanding tool pool. It derives numerous complex workflows from real business tasks and retains high-quality training trajectories through dual filtering. The system covers scenarios like data analysis, cross-system integration, and batch document processing, supporting task execution across more than 10 steps.

On the training side, the team moved away from pure supervised fine-tuning and adopted large-scale Agentic reinforcement learning. They used Harness Scaling to train across multiple mainstream Agent frameworks, preventing overfitting to a single interaction format. An asymmetric PPO architecture was introduced to address credit assignment in long-term tasks. A hierarchical reward mechanism was designed (core task results + behavioral norm constraints + failure exploration incentives) to balance effectiveness and robustness. The model also employs MOPD multi-teacher online strategy distillation to integrate capabilities from five expert models: long-term engineering, general Agentic, terminal usage, front-end aesthetics, and general knowledge. As a result, a single model can now handle multiple scenarios—writing code, running workflows, and generating front-end pages—without needing to switch.
Official evaluation data reveals: in code engineering, the SWE-Bench Pro score is 65.2, and the internal KAT Code Bench score is 53.1, enabling direct handling of complete Issues without manual decomposition. For Agentic tasks, the PinchBench score is 94.2, and the internal KAT Claw Bench score is 85.5, demonstrating excellent full-process stability.
KAT-Coder-Pro V2.5 is now fully available on the StreamLake platform (streamlake.com). It is open for API applications and technical documentation access, and the team has also released technical reports and developer exchange groups.
URL: https://streamlake.com/product/kat-coder
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation
Swedish AI Startup Lovable Eyes $13.2 Billion Valuation After Major Funding Round
As AI-driven coding tools gain traction, Swedish startup Lovable has secured a major funding round. The company aims to raise $3 billion, potentially boosting its valuation to $13.2 billion—double the $6.6 billion recorded last December. Menlo Ventur











