AI
AREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 28/09/2026, 20:00:00 UTC
Briefing created: 01/10/2026, 10:19:33 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 04:18:24 UTC.
Briefings
AREX-2 is a self-improving agent framework leveraging long-horizon reflective tasks to enhance LLM capabilities. Trained on supervised data from ML and programming domains, the agent uses Qwen3.8-27B to achieve strong performance on benchmarks like MLE-bench Lite (81.8) and Frontier-CS (70.7). It demonstrates sustained improvement across diverse tasks including DeepSearchQA (93.8) and BrowseComp (84.0) as its iteration budget expands.
Key points
- AREX-2 synthesizes long-horizon improvement trajectories from machine learning and algorithmic programming tasks to enable iterative refinement of LLM solutions.
- The agent trained on Qwen3.8-27B achieves high performance on benchmarks such as MLE-bench Lite (81.8) and Frontier-CS (70.7) while maintaining gains as its round budget increases.
Context and limitations
Results are limited to the specific benchmarks tested; performance on other domains remains unverified. Long-horizon trajectories were synthesized from existing data rather than generated autonomously. Some metrics like HLE and GAIA scores are reported but context on their methodology is not fully detailed.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 10:19:33 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeihewpdxblzwcj2bcz6eiucvvxzmquetudgjrp4cecjpb6dbdm44b4