# BAAI releases AREX-2, a 27b agent model

Digest AI · Generative AI & Models · published 2026-09-30T08:21:18Z

Canonical: https://digestai.news/story/baai-releases-arex-2-a-27b-agent-model

## Summary

baai has announced arex-2, a 27b-parameter long‑horizon agent model that builds on the qwen3.8 architecture. the model is trained on machine‑learning and algorithmic‑programming tasks with verifiable feedback, and it can iteratively refine solutions through propose‑measure‑reflect‑revise cycles.

arex‑2 supports a context length of 262,144 tokens and is evaluated on algorithmic programming, machine‑learning engineering, deep research, and general agentic reasoning. the paper reports performance on the frontier‑cs 188‑task agent track and on the mle‑lite any medal benchmark.

the model is released under the apache license 2.0 for research use and can be accessed via the transformers library with the model id "baai/arex‑2".

## Key points

- arex‑2 is a 27b‑parameter agent model based on qwen3.8
- context length is 262,144 tokens
- trained on coding, machine‑learning, and deep‑research tasks

## Why it matters

arex‑2 demonstrates how large multimodal models can perform long‑horizon reasoning and self‑improvement, potentially advancing research in iterative problem solving and deep‑research automation.

## Sources

1. [BAAI/AREX-2 - 27B - Agent model based on Qwen3.8 27B](https://huggingface.co/BAAI/AREX-2) (huggingface.co, 2026-09-30, primary source)

## Cite

Digest AI, "BAAI releases AREX-2, a 27b agent model", 30 September 2026, https://digestai.news/story/baai-releases-arex-2-a-27b-agent-model

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/baai-releases-arex-2-a-27b-agent-model.json