Kuaishou drops KAT-Coder Pro V2 for agentic coding
Kuaishou's KwaiPilot team has released KAT-Coder Pro V2, a 72B active parameter Mixture-of-Experts (MoE) model optimized for autonomous software engineering. Built with agentic reinforcement learning, the model features a 256K context window and achieves a 73.4% solve rate on SWE-Bench Verified, placing it in direct competition with frontier models like Claude 4.5 and GPT-5.
KAT-Coder Pro V2 is a category-defining challenger that matches frontier-level coding performance at a fraction of the cost. With a 73.4% SWE-Bench Verified solve rate and a 72B parameter Mixture-of-Experts architecture, it provides a superior balance of speed and reasoning depth. The model features native support for MCP and agentic frameworks, alongside a unique Web Aesthetics Generation feature for production-ready frontend design. Aggressive pricing (~$0.30/$1.20 per 1M tokens) significantly undercuts established competitors.
DISCOVERED
60d ago
2026-03-30
PUBLISHED
60d ago
2026-03-30
RELEVANCE
AUTHOR
Bijan Bowen
