{"slug": "arex-2-advancing-self-improving-agents-through-long-horizon-reflective-tasks", "title": "AREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks", "summary": "AREX-2, an LLM agent built on Qwen3.8-27B and trained on synthesized long-horizon improvement trajectories from machine learning and algorithmic programming tasks, scored 81.8 on MLE-bench Lite and 70.7 on Frontier-CS, according to the arXiv paper 2609.38288v1. The agent transferred to deep research benchmarks with 84.0 on BrowseComp, 52.6 on HLE, 92.2 on GAIA, and 93.8 on DeepSearchQA, and continued improving as its budget of iteration rounds grew. The authors conclude that long-horizon reflective data is an effective route toward self-improving agents.", "body_md": "arXiv:2609.38288v1 Announce Type: new \nAbstract: We present AREX-2, an effort to advance the self-improving capability of LLM agents, which we define as the ability to iteratively refine a solution at test time. This ability rests on two complementary capabilities: reflection, which produces a solution better than the current one, and long-horizon execution, which keeps the iteration effective over many rounds. We hypothesize that both capabilities are domain-agnostic, and can therefore be learned in scenarios that are well suited for supervision. Accordingly, we synthesize long-horizon improvement trajectories from machine learning and algorithmic programming tasks, two domains that offer verifiable feedback and reward sustained iteration. Trained on this data, our agent, built on Qwen3.8-27B, achieves strong results on MLE-bench Lite (81.8) and Frontier-CS (70.7), transfers to deep research with 84.0 on BrowseComp, 52.6 on HLE, 92.2 on GAIA, and 93.8 on DeepSearchQA, and keeps improving as its budget of rounds grows. These results show that long-horizon reflective data is an effective route toward self-improving agents.", "url": "https://wpnews.pro/news/arex-2-advancing-self-improving-agents-through-long-horizon-reflective-tasks", "canonical_source": "https://arxiv.org/abs/2609.38288", "published_at": "2026-10-01 04:00:00+00:00", "updated_at": "2026-10-01 04:17:17.622349+00:00", "lang": "en", "topics": ["ai-agents", "large-language-models", "ai-research", "machine-learning", "artificial-intelligence"], "entities": ["AREX-2", "Qwen3.8-27B", "MLE-bench Lite", "Frontier-CS", "BrowseComp", "HLE", "GAIA", "DeepSearchQA"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/arex-2-advancing-self-improving-agents-through-long-horizon-reflective-tasks", "markdown": "https://wpnews.pro/news/arex-2-advancing-self-improving-agents-through-long-horizon-reflective-tasks.md", "text": "https://wpnews.pro/news/arex-2-advancing-self-improving-agents-through-long-horizon-reflective-tasks.txt", "jsonld": "https://wpnews.pro/news/arex-2-advancing-self-improving-agents-through-long-horizon-reflective-tasks.jsonld"}}