HarborRL: An Open-Source Framework for Agentic Reinforcement Learning

·2026.10.27 22:31

Key point

The framework integrates model serving, trajectory generation, and policy updates into a single reproducible pipeline supporting Qwen and GLM models.

HarborRL is an open-source framework designed to simplify agentic reinforcement learning by unifying model serving, trajectory generation, tool execution, verifier rewards, and policy updates into one pipeline.
Supports Qwen and GLM models using GRPO and DAPO reinforcement learning algorithms.
Includes over 200 tasks such as Terminal-Bench, SWE-Bench, and Deep-SWE.
Compatible with more than 40 agent frameworks including Claude Code, Codex, LangGraph, and SWE-agent.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.