{"paper":{"title":"Breaking $\\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"GCPO replaces individual rollout competition with team-level credit assignment based on contributions to collective solution coverage.","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Haoxuan Chen, Jian-Fang Hu, Tianming Liang, Wei-Shi Zheng","submitted_at":"2026-05-12T03:20:24Z","abstract_excerpt":"Reinforcement learning with verifiers (RLVR) has become a central paradigm for improving LLM reasoning, yet popular group-based optimization algorithms like GRPO often suffer from exploration collapse, where the models prematurely converge on a narrow set of high-scoring patterns, lacking the ability to explore new solutions. Recent efforts attempt to alleviate this by adding entropy regularization or diversity bonus. However, these approaches do not change the \\textit{winner-takes-all} nature, where rollouts still compete for individual advantage rather than cooperating for maximizing global "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments across multiple reasoning benchmarks demonstrate that GCPO significantly improves both reasoning accuracy and solution diversity over existing approaches.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the determinant volume over reward-weighted semantic embeddings provides a reliable, unbiased measure of non-redundant solution coverage whose marginal contributions can be computed and redistributed without introducing new optimization pathologies or sensitivity to embedding choice.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GCPO shifts RLVR from rollout competition to team cooperation by assigning advantages via marginal contributions to a determinant-based coverage volume over semantic embeddings, yielding higher accuracy and solution diversity on reasoning benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"GCPO replaces individual rollout competition with team-level credit assignment based on contributions to collective solution coverage.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"811f064361e94f8e95068ab133f6c15bb946d0a481d6252b59f8c4d2b29ac605"},"source":{"id":"2605.11461","kind":"arxiv","version":2},"verdict":{"id":"167135f8-3676-4243-a715-400957f35937","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T01:53:17.683923Z","strongest_claim":"Experiments across multiple reasoning benchmarks demonstrate that GCPO significantly improves both reasoning accuracy and solution diversity over existing approaches.","one_line_summary":"GCPO shifts RLVR from rollout competition to team cooperation by assigning advantages via marginal contributions to a determinant-based coverage volume over semantic embeddings, yielding higher accuracy and solution diversity on reasoning benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the determinant volume over reward-weighted semantic embeddings provides a reliable, unbiased measure of non-redundant solution coverage whose marginal contributions can be computed and redistributed without introducing new optimization pathologies or sensitivity to embedding choice.","pith_extraction_headline":"GCPO replaces individual rollout competition with team-level credit assignment based on contributions to collective solution coverage."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.11461/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T12:35:47.127139Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T10:01:16.385739Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:27:15.467797Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"f3a30e2b5c553c5c8855fc2177d622744c629cc9279bc9fc569c646f70227da7"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"4d92269dc825e89728ba0dff14924c146e4ad67d8bb07e132ca431858ce1a53a"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}