Back to feed
arXiv cs.CL·

EnvFactory: Scaling Tool-Use Agents via Executable Environments Synthesis and Robust RL

Signal
78
Hype
25
In three linesEnvFactory automates creation of executable environments and synthesis of multi-turn trajectories for Agentic RL training. Using 85 verified environments across 7 domains, the framework generates 2,575 SFT/RL trajectories and improves Qwen3-series models by +15% on BFCLv3, +8.6% on MCP-Atlas, and +6% on conversational benchmarks.
Read source
Your take?
AI AgentsReinforcement learningCode generationBenchmarksPapers

Summary generated by Claude — human-verified