Which LLM + Harness? The Best Combinations for AI

Your AI Agent Failed. Blame the LLM or the Harness? When Agents Fail: Optimize your LLM or is your harness the main obstacle? New answers, especially for open-weight models. Kimi, Qwen, DeepSeek, and more. OpenCode, Codex, Claude Code, and more. All rights w/ authors: When Agents Fail: Model or Harness—Who’s to Blame? Shao Zhang, Hangfan Zhang, Yuanyuan Shi, Hao Li, Min-Le Su, Kangcong Li, Jingping Wang, Chunjiang Mu, Chen Zhang, Yang Chen, Bo Zhang, Shuyue Hu Shanghai Artificial Intelligence Laboratory #harnessengineering #aitechnology #aitech