Reproduce “Self-Credit: Credit Assignment for Long-Horizon Agentic RL via Intra-group Rescaling” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Fangcong Yin, Ming Yin, Tanvi Bhandarkar, Aashu Singh, Eryk Helenowski, Shlok Kumar Mishra, Saghar Hosseini, Karthik Abinav Sankararaman and 2 more

Self-Credit trains long-horizon search agents by using citations in their own summaries to assign turn-level learning signals. It matches similar cited actions across successful and failed attempts at the same question, then adjusts their training advantages according to outcome patterns. On BrowseComp-Plus, Qwen3-32B reaches 47.0% Pass@1, versus 41.4% for ECHO; citations are signals, not verified causal labels.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.