Microsoft’s OpenForgeRL trains harness-native agents end to end in any environment
Fighterdan · hf · 2026-07-25
Microsoft researchers introduce OpenForgeRL, an open-source framework for training harness-native agents end to end in any environment.
What it does
- Adds a lightweight proxy to record harness model calls as RL training data.
- Uses a Kubernetes orchestrator so each rollout runs in its own remote container.
- Lets researchers train agents inside real harnesses such as Claude Code, Codex, OpenClaw, and GUI/browser/computer-use setups.
Reported results
- OpenForgeClaw: 31.7 pass^3 on ClawEval and 55.9 pass@3 on ClawEval, plus 33.7 on QwenClawBench.
- OpenForgeGUI: 37.7 on OSWorld-Verified, 63.0 on Online-Mind2Web, and 72.3 on WebVoyager.
- The paper says these models beat similarly sized open baselines on nearly all benchmarks, and in GUI tasks can match or exceed models several times larger.
Takeaways
- Harness choice materially changes agent behavior.
- RL improves reliability, self-verification, tool coverage, and multi-step plan completion.
- Error recovery is still weak.
Related event: Microsoft Open-Sources OpenForgeRL for End-to-End Agent Training(4 posts)→
More from coding & agent
- Andrew Ng’s OpenWorker open beta runs local or cloud models and plugs into Slack — Saboo_Shubham_ · 2026-07-26
- Grok 4.5 tops several coding-agent benchmarks while claiming a lower cost per fix — XFreeze · 2026-07-26
- Recursive self-improvement may start with AI improving the code that builds its successors — imjustnewatai · 2026-07-26
- Agent inference costs are exploding as multi-agent workflows burn through tokens — 机器之心 · 2026-07-26
- Parallel scout agents can improve context understanding while cutting latency — pvncher · 2026-07-26
- Someone built a full 3D Doom-style game with SQL queries — thursdai_pod · 2026-07-26