{
  "id": 4483,
  "url": "https://arxiv.org/abs/2605.11853v2",
  "title": "GEAR: Granularity-Adaptive Advantage Reweighting for LLM Agents via Self-Distillation",
  "summary": "Reinforcement learning has become a widely used post-training approach for LLM agents, where training commonly relies on outcome-level rewards that provide only coarse supervision. While finer-grained credit assignment is promising for effective policy updates, obtaining reliable local credit and assigning it to the right parts of the long-horizon trajectory remains an open challenge. In this paper, we propose Granularity-adaptivE Advantage Reweighting (GEAR), an adaptive-granularity credit assi",
  "authors": "Sijia Li, Yuchen Huang, Zifan Liu, Yanping Li, Jingjing Fu, Li Zhao et al.",
  "category": "research",
  "topics": "regulation,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-12T09:38:38.000Z",
  "fetched_at": "2026-07-14T16:31:03.577Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/4483",
  "original_url": "https://arxiv.org/abs/2605.11853v2",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}