← sources

BAAI/AREX-2 - 27B - Agent model based on Qwen3.8 27B

source: reddit (reddit:LocalLLaMA)published 2026-09-30 08:21ingested 2026-10-01 02:44

retained text

"AREX-2 is a 27B-parameter long-horizon agent model from the Beijing Academy of Artificial Intelligence (BAAI). It learns to improve a solution over multiple test-time rounds: propose, measure, reflect, and revise.

AREX-2 is trained on machine-learning and algorithmic-programming tasks with verifiable feedback, together with the existing AREX deep-research data. The learned self-improvement behavior transfers to deep research without adding new search trajectories.

* **Architecture:** Dense Qwen3.8-compatible multimodal model
* **Parameters:** 27B
* **Context length:** 262,144 tokens

# [](https://huggingface.co/BAAI/AREX-2#key-features)Key features

* **Long-horizon self-improvement:** turns extra test-time rounds into useful solution refinement.
* **Feedback-driven reflection:** reads scores, logs, errors, and timings to decide what to change next.
* **Cross-domain performance:** training on coding and machine-learning tasks also improves the model's deep-research performance.
* **Long-horizon reasoning:** sustains productive iteration as the task budget grows."

Gguf's - [https://huggingface.co/mradermacher/AREX-2-GGUF](https://huggingface.co/mradermacher/AREX-2-GGUF)

document fetched 2026-10-01 00:46 · canonical https://reddit.com/r/LocalLLaMA/comments/1wtzedu/baaiarex2_27b_agent_model_based_on_qwen38_27b