Ant Group Jointly Releases Stable Version of Open-Source Reinforcement Learning Training Framework AReaL v1.0 with Tsinghua University

K-LinePoet · 2026-03-23T12:38:03+00:00

Ant Group and Tsinghua University released the stable version of AReaL v1.0, an open-source reinforcement learning training framework, featuring "one-click access to RL training." The framework supports multiple agents and requires no code modifications. Through the Proxy Worker relay layer, it enables zero-modification integration, making it convenient for developers to optimize decision-making.

K-LinePoet

2026-03-23 12:38:03

Abstract generation in progress

On March 4th, Ant Group, in collaboration with Tsinghua University, released the stable version of the open-source reinforcement learning training framework AReaL v1.0. This version features “Agent one-click access to RL training”: no code modifications needed, compatible with various agent frameworks, making reinforcement learning training for intelligent agents ready to use out of the box. AReaL is the first fully asynchronous, decoupled training and inference large model reinforcement learning system, enabling agents to receive feedback and continuously optimize decisions through real task interactions. The release of version 1.0 makes zero-modification access to RL training for any agent a reality—by adding a Proxy Worker relay layer between the agent and the training system, developers only need to change a request address to connect. (Science and Technology Innovation Board Daily)

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.