{"paper":{"title":"Reinforcement Learning for Reasoning in Large Language Models with One Training Example","license":"http://creativecommons.org/licenses/by/4.0/","headline":"One training example via reinforcement learning lifts an LLM's math reasoning score from 36% to 74% on MATH500.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Baolin Peng, Hao Cheng, Jianfeng Gao, Kuan Wang, Liliang Ren, Liyuan Liu, Qing Yang, Shuohang Wang, Simon Shaolei Du, Weizhu Chen, Xuehai He, Yelong Shen, Yiping Wang, Zhiyuan Zeng","submitted_at":"2025-04-29T09:24:30Z","abstract_excerpt":"We show that reinforcement learning with verifiable reward using one training example (1-shot RLVR) is effective in incentivizing the math reasoning capabilities of large language models (LLMs). Applying RLVR to the base model Qwen2.5-Math-1.5B, we identify a single example that elevates model performance on MATH500 from 36.0% to 73.6% (8.6% improvement beyond format correction), and improves the average performance across six common mathematical reasoning benchmarks from 17.6% to 35.7% (7.0% non-format gain). This result matches the performance obtained using the 1.2k DeepScaleR subset (MATH5"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Reinforcement learning with verifiable reward using one training example (1-shot RLVR) is effective in incentivizing the math reasoning capabilities of large language models (LLMs). Applying RLVR to the base model Qwen2.5-Math-1.5B, we identify a single example that elevates model performance on MATH500 from 36.0% to 73.6%.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the single chosen example is not specially selected in a way that inflates generalization, and that the observed gains arise from the RL policy gradient rather than incidental effects of the training setup or prompt format.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"One training example via RLVR boosts LLM math reasoning from 17.6% to 35.7% average across six benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"One training example via reinforcement learning lifts an LLM's math reasoning score from 36% to 74% on MATH500.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"36735173058404e2fa8e7aad18c2de165a2691a58c91a23b42e5aca735632a56"},"source":{"id":"2504.20571","kind":"arxiv","version":3},"verdict":{"id":"c7b86817-aa42-4484-a016-6876f0248f25","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T19:47:48.868083Z","strongest_claim":"Reinforcement learning with verifiable reward using one training example (1-shot RLVR) is effective in incentivizing the math reasoning capabilities of large language models (LLMs). Applying RLVR to the base model Qwen2.5-Math-1.5B, we identify a single example that elevates model performance on MATH500 from 36.0% to 73.6%.","one_line_summary":"One training example via RLVR boosts LLM math reasoning from 17.6% to 35.7% average across six benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the single chosen example is not specially selected in a way that inflates generalization, and that the observed gains arise from the RL policy gradient rather than incidental effects of the training setup or prompt format.","pith_extraction_headline":"One training example via reinforcement learning lifts an LLM's math reasoning score from 36% to 74% on MATH500."},"references":{"count":69,"sample":[{"doi":"","year":2024,"title":"Learning to reason with llms","work_id":"170ab9a7-f02d-4838-a217-516dcecfaaf1","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","ref_index":2,"cited_arxiv_id":"2501.12948","is_internal_anchor":true},{"doi":"","year":2025,"title":"Kimi k1.5: Scaling Reinforcement Learning with LLMs","work_id":"bff96ab1-bd6a-4585-be23-74fdb51969c7","ref_index":3,"cited_arxiv_id":"2501.12599","is_internal_anchor":true},{"doi":"","year":2024,"title":"On designing effective rl reward at training time for llm reasoning","work_id":"5c46cc3b-db37-4655-a0ff-f28ab6c32a1f","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Tulu 3: Pushing Frontiers in Open Language Model Post-Training","work_id":"28c9dbea-056a-48c2-8000-85f809827e45","ref_index":5,"cited_arxiv_id":"2411.15124","is_internal_anchor":true}],"resolved_work":69,"snapshot_sha256":"57a5e9920dcc9893c568b300a373c6224244244cb7150ad9389f9bf9eeae1bad","internal_anchors":29},"formal_canon":{"evidence_count":1,"snapshot_sha256":"dbbdd7d28668dbf699e7d875b7e1415989d324785463f554cf12a2052ab56764"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}