InclusionAI Community to Present AReaL Reinforcement Learning Framework and AWorld Multi‑Agent Framework at ICLR 2025

The International Conference on Learning Representations (ICLR) 2025 will be held from April 24‑28 at Singapore EXPO, and the InclusionAI open‑source community, launched by Ant Group, will share the newest developments of its reinforcement‑learning framework AReaL and multi‑agent framework AWorld during the event.

AReaL (Ant Reasoning RL), co‑developed by Ant Technology Research Institute and Tsinghua University’s Institute for Interdisciplinary Information, released version 0.2 – AReaL‑boba – which achieves significant improvements in performance and cost efficiency. It supports large‑scale distributed training, enabling a 1.5 B parameter reasoning model to be trained on 128 H800 GPUs in one day, and a 7 B model on 256 H800 GPUs within two days, attaining state‑of‑the‑art scores on the AIME benchmarks.

Assistant Professor Wu Yi from Tsinghua University will reveal the “recipe” behind AReaL‑boba, sharing practical insights on building reasoning models.

AWorld (Agent World) is Ant’s early‑stage multi‑agent framework open‑sourced through the InclusionAI community, designed to let individuals and teams easily construct and customize multi‑agent AI assistants. It emphasizes collaborative mechanisms, environment construction, and real‑world problem solving, with recent progress demonstrated on the GAIA benchmark.

Gu Jin‑jie, head of Ant Group’s Agent project, will present the latest advances of AWorld, discussing industry observations and practical experiences with multi‑agent technology.

Since February, the InclusionAI community has focused on large language models, reinforcement learning, model training and inference, and other AGI‑related frameworks, releasing projects such as the Ling language model, Ring inference model, AReal, AWorld, and PromptCoT, all under principles of fairness, transparency, and collaboration.

ICLR, founded by leading deep‑learning pioneers Yoshua Bengio and Yann LeCun, gathers researchers, engineers, entrepreneurs, and students to advance representation learning, a core subfield of artificial intelligence.

Attendees are invited to meet the InclusionAI team in Singapore on April 24.