AREX-2 trains a 27B model with long-horizon reflection
The AREX research group released AREX-2, a model based on Qwen3.8-27B and trained with long-horizon reflection data for machine-learning and algorithm-programming tasks. The team reported scores of 81.8 on MLE-bench Lite, 70.7 on Frontier-CS, 84.0 on BrowseComp and 52.6 on HLE. It also said the training helped capabilities transfer to deep research and scientific problem-solving. The report presents long-horizon reflection as a trainable and verifiable approach to recursive self-improvement, which is the team’s research interpretation rather than established proof of that capability.
AREX-2 provides open weights, a paper and details of a reflection-and-feedback training process. Its reported results offer a practical case for training a 27B model to use feedback, revise solutions and sustain iterative improvement across tasks.
机器之心 2026-10-27 12:03 北京
AI自我改进有了训练路径。
机器之心发布
当模型学会持续迭代改进算法，便推开了通往智能自主进化的大门。近日，由来自智源研究院、中国科技大学、中国人民大学等多个机构的研究人员组成的 AREX 研究小组发布 AREX-2 模型。该模型以轻量级的 Qwen3.8-27B 为基座，基于围绕机器学习与算法编程场景构建的长程反思数据训练而成，不仅在 MLE-bench Lite、Frontier-CS 上展现出领先的机器学习与算法编程能力，更将这种能力迁移至深度研究与科学问题求解，在 BrowseComp、HLE 等重要基准上刷新了开源轻量模型的最佳表现。这些结果表明，长程反思训练有望帮助模型习得跨领域的自我改进能力，让更多测试时投入持续转化为更好的结果，为 RSI 提供一条可训练、可验证的实现路径。
#AREX#AREX-2#Qwen3.8-27B#BAAI#University of Science and Technology of China#Renmin University of China#GPT-5.6 Sol#Kimi-K3#Frontis-MA1-35B#Kimi-K2.6#DeepSeek-V4-Pro