Dockerless is an agentic verifier for coding-agent patches that avoids executing candidate code in Docker or repository-specific environments. It explores the repository and gathers evidence to judge patch correctness instead of relying only on reference-patch matching. On a verifier evaluation benchmark, it reportedly beats the strongest open-source verifier by 14.3 AUC points. Used for both SFT trajectory filtering and RL rewards, Dockerless enables an environment-free post-training pipeline. The resulting model achieves resolve rates of 62.0% on SWE-bench Verified, 50.0% on Multilingual, and 35.2% on Pro, exceeding the Qwen3.5-9B baseline by 2.4, 8.7, and 2.9 points.
No heat snapshots are available in the last 24 hours.