{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ZXEVIDXGJBUBRIEHXKYBEA5T7O","short_pith_number":"pith:ZXEVIDXG","schema_version":"1.0","canonical_sha256":"cdc9540ee6486818a087bab01203b3fb9b97a97f76415f84929b473329a2656f","source":{"kind":"arxiv","id":"2503.18069","version":1},"attestation_state":"computed","paper":{"title":"Long Is More Important Than Difficult for Training Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chang Liu, Danhao Zhu, Fei Huang, Si Shen, Tiansheng Zheng, Zhixiao Zhao","submitted_at":"2025-03-23T13:33:59Z","abstract_excerpt":"Difficult problems, which often result in long reasoning traces, are widely recognized as key factors for enhancing the performance of reasoning models. However, such high-challenge problems are scarce, limiting the size of available datasets. In this paper, we propose a simple method to decouple the reliance on problem difficulty. First, we empirically demonstrate that reasoning length, rather than problem difficulty, primarily influences the performance of trained models. Second, we identify a scaling law on reasoning length, showing that model performance increases in a log-linear fashion a"},"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":"2503.18069","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-03-23T13:33:59Z","cross_cats_sorted":[],"title_canon_sha256":"0ea03e7f688a382aa65a448db423fb3d3e58fc2413c560aeef71da75097c1689","abstract_canon_sha256":"4179759b2a426f61e31e87a47dfcbc52b1098d7e2503da22a12496d96ccfc41e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:38:01.912365Z","signature_b64":"oXU4Q+aCv+SNfIfSzart0X00EoEumwmZsStYLgz4BytmFDVblseMc3u1XRDB/HFOPHY2UaZ+IER+5qZip1P2Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cdc9540ee6486818a087bab01203b3fb9b97a97f76415f84929b473329a2656f","last_reissued_at":"2026-07-05T10:38:01.911944Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:38:01.911944Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Long Is More Important Than Difficult for Training Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chang Liu, Danhao Zhu, Fei Huang, Si Shen, Tiansheng Zheng, Zhixiao Zhao","submitted_at":"2025-03-23T13:33:59Z","abstract_excerpt":"Difficult problems, which often result in long reasoning traces, are widely recognized as key factors for enhancing the performance of reasoning models. However, such high-challenge problems are scarce, limiting the size of available datasets. In this paper, we propose a simple method to decouple the reliance on problem difficulty. First, we empirically demonstrate that reasoning length, rather than problem difficulty, primarily influences the performance of trained models. Second, we identify a scaling law on reasoning length, showing that model performance increases in a log-linear fashion a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.18069","kind":"arxiv","version":1},"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/2503.18069/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":"2503.18069","created_at":"2026-07-05T10:38:01.912004+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.18069v1","created_at":"2026-07-05T10:38:01.912004+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.18069","created_at":"2026-07-05T10:38:01.912004+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZXEVIDXGJBUB","created_at":"2026-07-05T10:38:01.912004+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZXEVIDXGJBUBRIEH","created_at":"2026-07-05T10:38:01.912004+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZXEVIDXG","created_at":"2026-07-05T10:38:01.912004+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2508.05928","citing_title":"Mitigating Think-Answer Mismatch in LLM Reasoning Through Noise-Aware Advantage Reweighting","ref_index":16,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZXEVIDXGJBUBRIEHXKYBEA5T7O","json":"https://pith.science/pith/ZXEVIDXGJBUBRIEHXKYBEA5T7O.json","graph_json":"https://pith.science/api/pith-number/ZXEVIDXGJBUBRIEHXKYBEA5T7O/graph.json","events_json":"https://pith.science/api/pith-number/ZXEVIDXGJBUBRIEHXKYBEA5T7O/events.json","paper":"https://pith.science/paper/ZXEVIDXG"},"agent_actions":{"view_html":"https://pith.science/pith/ZXEVIDXGJBUBRIEHXKYBEA5T7O","download_json":"https://pith.science/pith/ZXEVIDXGJBUBRIEHXKYBEA5T7O.json","view_paper":"https://pith.science/paper/ZXEVIDXG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.18069&json=true","fetch_graph":"https://pith.science/api/pith-number/ZXEVIDXGJBUBRIEHXKYBEA5T7O/graph.json","fetch_events":"https://pith.science/api/pith-number/ZXEVIDXGJBUBRIEHXKYBEA5T7O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZXEVIDXGJBUBRIEHXKYBEA5T7O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZXEVIDXGJBUBRIEHXKYBEA5T7O/action/storage_attestation","attest_author":"https://pith.science/pith/ZXEVIDXGJBUBRIEHXKYBEA5T7O/action/author_attestation","sign_citation":"https://pith.science/pith/ZXEVIDXGJBUBRIEHXKYBEA5T7O/action/citation_signature","submit_replication":"https://pith.science/pith/ZXEVIDXGJBUBRIEHXKYBEA5T7O/action/replication_record"}},"created_at":"2026-07-05T10:38:01.912004+00:00","updated_at":"2026-07-05T10:38:01.912004+00:00"}