KAT-Coder-Pro V1 is KwaiPilot's agentic coding model. It achieves a 73.4% resolve rate on SWE-Bench Verified with a context window of 256K tokens, parallel tool calling, and multi-turn support.
import { streamText } from 'ai'
const result = streamText({ model: 'kwaipilot/kat-coder-pro-v1', prompt: 'Why is the sky blue?'})KAT-Coder-Pro V1 is the first production release from KwaiPilot's KAT-Coder series, designed for agentic software engineering at repository scale. On SWE-Bench Verified, a standard benchmark for autonomous issue resolution, it achieves a 73.4% resolve rate.
The model's agentic design centers on two capabilities. Parallel tool calling lets the model issue multiple tool calls simultaneously instead of waiting sequentially. Multi-turn support enables sustained coding sessions. That efficiency gain matters for agent loops billed per token or per call.
KAT-Coder-Pro V1 covers eight task types: feature implementation, feature enhancement, bug fixing, refactoring, performance optimization, test case generation, code understanding, and configuration and deployment. The context window of 256K tokens lets it ingest large codebases, multi-file diffs, and extended conversation history in a single pass. Full technical details appear at https://novita.ai.