# A 27B Open-Weight Agent From BAAI Posts 81.8 on MLE-bench Lite by Learning to Iterate

AREX-2 trains on synthesized long-horizon improvement trajectories and transfers to deep research, reaching 84.0 on BrowseComp and 52.6 on Humanity's Last Exam.

- Published: 2026-10-01T06:20:51.871Z
- Canonical: https://polylog.news/ai/2026-10-01/a-27b-open-weight-agent-from-baai-posts-81-8-on-mle-bench-li
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.AI](https://arxiv.org/abs/2609.38288), [arXiv cs.AI](https://arxiv.org/abs/2609.38294)

A team from the Beijing Academy of Artificial Intelligence (BAAI) posted AREX-2, which decomposes test-time self-improvement into two learnable capabilities: reflection, which produces a solution better than the current one, and long-horizo…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-10-01/a-27b-open-weight-agent-from-baai-posts-81-8-on-mle-bench-li (subscription information: https://polylog.news/pricing).