{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ZCMT7GW5BDOK5Q63FTI4NMGATG","short_pith_number":"pith:ZCMT7GW5","schema_version":"1.0","canonical_sha256":"c8993f9add08dcaec3db2cd1c6b0c099b14ba68d11e92d0d8b997cd404f1695a","source":{"kind":"arxiv","id":"2505.07773","version":4},"attestation_state":"computed","paper":{"title":"Agent RL Scaling Law: Agent RL with Spontaneous Code Execution for Mathematical Problem Solving","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Haotian Xu, Jian Hu, Weinong Wang, Wenqiang Zhang, Xing W, Xinji Mai, Yingying Zhang, Zhong-Zhi Li","submitted_at":"2025-05-12T17:23:34Z","abstract_excerpt":"Large Language Models (LLMs) often struggle with mathematical reasoning tasks requiring precise, verifiable computation. While Reinforcement Learning (RL) from outcome-based rewards enhances text-based reasoning, understanding how agents autonomously learn to leverage external tools like code execution remains crucial. We investigate RL from outcome-based rewards for Tool-Integrated Reasoning, ZeroTIR, training base LLMs to spontaneously generate and execute Python code for mathematical problems without supervised tool-use examples. Our central contribution is we demonstrate that as RL trainin"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2505.07773","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-05-12T17:23:34Z","cross_cats_sorted":[],"title_canon_sha256":"2a6c3dcfd1ce1ac75f768aa109db9618d58ca8349001ce39b6355e92950efd51","abstract_canon_sha256":"fe15f98a87556bcc425d997dca362c7f7b837734165da3a753d414be7e51b8eb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:56:19.556205Z","signature_b64":"Qs3JKQW2QzXev0Mmk2x9ipr7tEAHYXpZSOG+xJ2xeATIMzNWVroDWc3TGUK7gpj8D9LWP/X+s4rlYC1pl/tpAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c8993f9add08dcaec3db2cd1c6b0c099b14ba68d11e92d0d8b997cd404f1695a","last_reissued_at":"2026-07-05T11:56:19.555718Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:56:19.555718Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Agent RL Scaling Law: Agent RL with Spontaneous Code Execution for Mathematical Problem Solving","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Haotian Xu, Jian Hu, Weinong Wang, Wenqiang Zhang, Xing W, Xinji Mai, Yingying Zhang, Zhong-Zhi Li","submitted_at":"2025-05-12T17:23:34Z","abstract_excerpt":"Large Language Models (LLMs) often struggle with mathematical reasoning tasks requiring precise, verifiable computation. While Reinforcement Learning (RL) from outcome-based rewards enhances text-based reasoning, understanding how agents autonomously learn to leverage external tools like code execution remains crucial. We investigate RL from outcome-based rewards for Tool-Integrated Reasoning, ZeroTIR, training base LLMs to spontaneously generate and execute Python code for mathematical problems without supervised tool-use examples. Our central contribution is we demonstrate that as RL trainin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.07773","kind":"arxiv","version":4},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2505.07773/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2505.07773","created_at":"2026-07-05T11:56:19.555790+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.07773v4","created_at":"2026-07-05T11:56:19.555790+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.07773","created_at":"2026-07-05T11:56:19.555790+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZCMT7GW5BDOK","created_at":"2026-07-05T11:56:19.555790+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZCMT7GW5BDOK5Q63","created_at":"2026-07-05T11:56:19.555790+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZCMT7GW5","created_at":"2026-07-05T11:56:19.555790+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":13,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24233","citing_title":"Latent Visual States for Efficient Multimodal Reasoning","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19236","citing_title":"STARE: Surprisal-Guided Token-Level Advantage Reweighting for Policy Entropy Stability","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2606.13316","citing_title":"ReSum: Synergizing LLM Reasoning and Summarization with Reinforcement Learning","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11652","citing_title":"IAPO: Input Attribution-Aware Policy Optimization for Tool Use in Small Multimodal Agents","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27859","citing_title":"Rethinking Agentic Reinforcement Learning In Large Language Models","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2510.00568","citing_title":"ReSeek: A Self-Correcting Framework for Search Agents with Instructive Rewards","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2509.07969","citing_title":"Mini-o3: Scaling Up Reasoning Patterns and Interaction Turns for Visual Search","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2511.02734","citing_title":"CostBench: Evaluating Multi-Turn Cost-Optimal Planning and Adaptation in Dynamic Environments for LLM Tool-Use Agents","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2603.16876","citing_title":"Multi-Modal Multi-Agent Reinforcement Learning for Radiology Report Generation","ref_index":81,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27859","citing_title":"Rethinking Agentic Reinforcement Learning In Large Language Models","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09931","citing_title":"PruneTIR: Inference-Time Tool Call Pruning for Effective yet Efficient Tool-Integrated Reasoning","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27859","citing_title":"Rethinking Agentic Reinforcement Learning In Large Language Models","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06777","citing_title":"Walk the Talk: Bridging the Reasoning-Action Gap for Thinking with Images via Multimodal Agentic Policy Optimization","ref_index":51,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZCMT7GW5BDOK5Q63FTI4NMGATG","json":"https://pith.science/pith/ZCMT7GW5BDOK5Q63FTI4NMGATG.json","graph_json":"https://pith.science/api/pith-number/ZCMT7GW5BDOK5Q63FTI4NMGATG/graph.json","events_json":"https://pith.science/api/pith-number/ZCMT7GW5BDOK5Q63FTI4NMGATG/events.json","paper":"https://pith.science/paper/ZCMT7GW5"},"agent_actions":{"view_html":"https://pith.science/pith/ZCMT7GW5BDOK5Q63FTI4NMGATG","download_json":"https://pith.science/pith/ZCMT7GW5BDOK5Q63FTI4NMGATG.json","view_paper":"https://pith.science/paper/ZCMT7GW5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.07773&json=true","fetch_graph":"https://pith.science/api/pith-number/ZCMT7GW5BDOK5Q63FTI4NMGATG/graph.json","fetch_events":"https://pith.science/api/pith-number/ZCMT7GW5BDOK5Q63FTI4NMGATG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZCMT7GW5BDOK5Q63FTI4NMGATG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZCMT7GW5BDOK5Q63FTI4NMGATG/action/storage_attestation","attest_author":"https://pith.science/pith/ZCMT7GW5BDOK5Q63FTI4NMGATG/action/author_attestation","sign_citation":"https://pith.science/pith/ZCMT7GW5BDOK5Q63FTI4NMGATG/action/citation_signature","submit_replication":"https://pith.science/pith/ZCMT7GW5BDOK5Q63FTI4NMGATG/action/replication_record"}},"created_at":"2026-07-05T11:56:19.555790+00:00","updated_at":"2026-07-05T11:56:19.555790+00:00"}