Tsinghua Releases VeriLoopCoder-E1: Open-Source Coding Agent with Evidence-Governed Self-Correction
机器之心 · wechat · 2026-08-02
A team from Tsinghua University Shenzhen International Graduate School released VeriLoopCoder-E1, an open-source code repair model. Built on Qwen3.6-27B, it achieved three first-place rankings among open-source models (≤32B) across four SWE benchmarks.
The model's core strength lies in the synergy between narrow-domain PEFT fine-tuning and the Self-Harness control stack. The team proposed the Evidence-Governed Spiral mechanism, upgrading mechanical retries into a closed loop of 'Evidence-Falsification-Exploration-Repair-Verification-Internalization'. This requires every generation and fix to be validated by evidence, allowing verified corrections to influence future error-discovery methods, pushing towards recursive self-improvement.
While the model weights are open-sourced under Apache 2.0, the Self-Harness control stack—responsible for multi-round task orchestration—is retained. The team will now focus on building general-purpose Code Agents capable of predicting and managing the consequences of their actions.
More from coding & agent
- Tencent Releases UI-Mate-27B, a Desktop GUI Agent Model — tencent · 2026-08-24
- Comparing AI Subscriptions: DeepSeek API vs. Claude Pro vs. Local LLMs — Unlikely_Bluejay5392 · 2026-08-24
- Claude Code introduces 'Remote Control' feature to boost coding efficiency — rohanpaul_ai · 2026-08-24
- rauchg lays out fx extension philosophy: MCP, Skills, Plugins and Unix composition — AccBalanced · 2026-08-24
- Netflix details its production LLM judge: hundreds of thousands of recommendations scored weekly — omarsar0 · 2026-08-24
- smolvm passes Simon Willison's Fable 5 agent test as a secure sandbox — yawnxyz · 2026-08-24