AI
AREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks
Publication: cs.AI updates on arXiv.org
Author: Hongjin Qian, Chaofan Li, Kun Luo, Wenqing Wei, Jianlyu Chen, Shuqi Lu, Yuyang Hu, Hongwang Xiao, Hui Wang, Chaozhuo Li, Qiwei Ye, Zhicheng Dou, Defu Lian, Zheng Liu
Source published: 01/10/2026, 04:00:00 UTC
Briefing created: 01/10/2026, 04:08:11 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 04:03:24 UTC.
Briefings
AREX-2 is a new LLM agent framework designed to advance self-improving capabilities by combining reflection and long-horizon execution. Trained on supervised data from ML and programming tasks, the agent achieves strong performance on benchmarks like MLE-bench Lite (81.8) and Frontier-CS (70.7), with results improving as its training budget increases. The study hypothesizes that these capabilities are domain-agnostic and can be learned in well-suited scenarios.
Key points
- AREX-2 is an LLM agent framework built on Qwen3.8-27B that combines reflection and long-horizon execution to improve self-improving capabilities.
- The agent achieves strong performance on benchmarks such as MLE-bench Lite (81.8) and Frontier-CS (70.7), with results improving as the training budget grows.
Context and limitations
Results are based on a single preprint from arXiv:2609.38288v1, which may not reflect final experimental outcomes. Long-horizon trajectories were synthesized from machine learning and algorithmic programming tasks, limiting generalizability to other domains.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 04:08:11 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeigtmjwtcfe4altenf2yt4yomuwyxg5hqxwzmrs37wddkejtcncic4