KAT-Coder-V2通过分领域训练与融合，实现高效智能编程，性能接近顶级模型。
KAT-Coder-V2 Technical Report
- 分五个专业领域独立训练后统一，提升代码生成能力
- 在SWE-bench上达79.6%，在PinchBench上超88.7%，表现优异
- 适合需要高精度编程的开发者与研究者使用
我们提出KAT-Coder-V2，由快手KwaiKAT团队开发的智能编程模型。该模型采用“专精-融合”范式，将智能编程分解为SWE、WebCoding、Terminal、WebSearch和General五个专家领域，分别进行监督微调与强化学习，再通过在线策略蒸馏整合为单一模型。我们构建了KwaiEnv，一个支持数万并发沙箱实例的模块化基础设施，并沿任务复杂度、意图对齐和支架泛化方向扩展强化学习训练。此外，提出MCLA稳定MoE强化学习训练，Tree Training消除树状轨迹中的冗余计算，最高提速6.2倍。KAT-Coder-V2在SWE-bench Verified上取得79.6%准确率（低于Claude Opus 4.6的80.8%），在PinchBench上达88.7%（超越GLM-5和MiniMax M2.7），在全部三个前端美学场景中排名第一，在Terminal-Bench Hard上保持46.8分，tau^2-Bench上达93.9分。模型已公开，可访问https://streamlake.com/product/kat-coder。
原文摘要 · Abstract (English)
We present KAT-Coder-V2, an agentic coding model developed by the KwaiKAT team at Kuaishou. KAT-Coder-V2 adopts a "Specialize-then-Unify" paradigm that decomposes agentic coding into five expert domains - SWE, WebCoding, Terminal, WebSearch, and General - each undergoing independent supervised fine-tuning and reinforcement learning, before being consolidated into a single model via on-policy distillation. We develop KwaiEnv, a modular infrastructure sustaining tens of thousands of concurrent sandbox instances, and scale RL training along task complexity, intent alignment, and scaffold generalization. We further propose MCLA for stabilizing MoE RL training and Tree Training for eliminating redundant computation over tree-structured trajectories with up to 6.2x speedup. KAT-Coder-V2 achieves 79.6% on SWE-bench Verified (vs. Claude Opus 4.6 at 80.8%), 88.7 on PinchBench (surpassing GLM-5 and MiniMax M2.7), ranks first across all three frontend aesthetics scenarios, and maintains strong generalist scores on Terminal-Bench Hard (46.8) and tau^2-Bench (93.9). Our model is publicly available at https://streamlake.com/product/kat-coder.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。