{"paper":{"title":"Deep Think with Confidence","license":"http://creativecommons.org/licenses/by/4.0/","headline":"DeepConf uses model-internal confidence to filter weak reasoning traces, reaching 99.9% accuracy on AIME 2025 while cutting tokens by 84.7%.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Jiawei Zhao, Xuewei Wang, Yichao Fu, Yuandong Tian","submitted_at":"2025-08-21T05:48:38Z","abstract_excerpt":"Large Language Models (LLMs) have shown great potential in reasoning tasks through test-time scaling methods like self-consistency with majority voting. However, this approach often leads to diminishing returns in accuracy and high computational overhead. To address these challenges, we introduce Deep Think with Confidence (DeepConf), a simple yet powerful method that enhances both reasoning efficiency and performance at test time. DeepConf leverages model-internal confidence signals to dynamically filter out low-quality reasoning traces during or after generation. It requires no additional mo"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"on challenging benchmarks such as AIME 2025, DeepConf@512 achieves up to 99.9% accuracy and reduces generated tokens by up to 84.7% compared to full parallel thinking.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That model-internal confidence signals reliably indicate the quality or correctness of individual reasoning traces, allowing effective dynamic filtering without additional training or tuning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DeepConf filters low-quality reasoning traces using model-internal confidence signals, reaching up to 99.9% accuracy on AIME 2025 while reducing generated tokens by up to 84.7% versus full parallel sampling.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"DeepConf uses model-internal confidence to filter weak reasoning traces, reaching 99.9% accuracy on AIME 2025 while cutting tokens by 84.7%.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5db56bd422fe8c58ed9b577bcb5bc786e62c63c983f183ad154af04a9e6a406d"},"source":{"id":"2508.15260","kind":"arxiv","version":1},"verdict":{"id":"7485b2b8-cca9-4e0f-ab75-e669a96c37ed","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T11:26:17.715523Z","strongest_claim":"on challenging benchmarks such as AIME 2025, DeepConf@512 achieves up to 99.9% accuracy and reduces generated tokens by up to 84.7% compared to full parallel thinking.","one_line_summary":"DeepConf filters low-quality reasoning traces using model-internal confidence signals, reaching up to 99.9% accuracy on AIME 2025 while reducing generated tokens by up to 84.7% versus full parallel sampling.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That model-internal confidence signals reliably indicate the quality or correctness of individual reasoning traces, allowing effective dynamic filtering without additional training or tuning.","pith_extraction_headline":"DeepConf uses model-internal confidence to filter weak reasoning traces, reaching 99.9% accuracy on AIME 2025 while cutting tokens by 84.7%."},"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"}