{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:4KJIW2UUNFPBFJVUWBL7XTIXHZ","short_pith_number":"pith:4KJIW2UU","schema_version":"1.0","canonical_sha256":"e2928b6a94695e12a6b4b057fbcd173e67049a69e83bf24e5a48b358115271c4","source":{"kind":"arxiv","id":"2606.25519","version":1},"attestation_state":"computed","paper":{"title":"Quantization Inflates Reasoning: Token Inflation as a Hidden Cost of Low-Bit Reasoning Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Beichen Huang, Li Zhang, Masahiro Tanaka, Minjia Zhang, Olatunji Ruwase, Walid Krichene, Xinyu Lian","submitted_at":"2026-06-24T07:54:55Z","abstract_excerpt":"Quantization is widely used to reduce the inference cost of large language models, but its effect on reasoning models is not fully captured by final-answer accuracy or per-token latency. We show that low-bit post-training quantization can introduce a hidden test-time compute cost: quantized reasoning models often generate longer chains of thought even when they still answer correctly. Across mathematical reasoning, code generation, scientific question answering, and agentic tool-use benchmarks, we find that INT4/INT3 quantization can preserve accuracy but increase reasoning-token usage, offset"},"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":"2606.25519","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-24T07:54:55Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"5f4cd7a6fa659e0de923f30e3893e03c99562740f770c7a5734a154b2fb4b0f4","abstract_canon_sha256":"5f3b425bf8144cafe924a7b2ca996d6da0190712fe8de67180976c791ce191de"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-25T01:18:07.703914Z","signature_b64":"HtybVFPSVo6oyXiH5rpgfAFDTWTDLKzXSvnllo6TPUGvYTDhwbXIY14rQNXJR8SBC4/ThJFnIzaC4EOcMaXPAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e2928b6a94695e12a6b4b057fbcd173e67049a69e83bf24e5a48b358115271c4","last_reissued_at":"2026-06-25T01:18:07.703477Z","signature_status":"signed_v1","first_computed_at":"2026-06-25T01:18:07.703477Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Quantization Inflates Reasoning: Token Inflation as a Hidden Cost of Low-Bit Reasoning Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Beichen Huang, Li Zhang, Masahiro Tanaka, Minjia Zhang, Olatunji Ruwase, Walid Krichene, Xinyu Lian","submitted_at":"2026-06-24T07:54:55Z","abstract_excerpt":"Quantization is widely used to reduce the inference cost of large language models, but its effect on reasoning models is not fully captured by final-answer accuracy or per-token latency. We show that low-bit post-training quantization can introduce a hidden test-time compute cost: quantized reasoning models often generate longer chains of thought even when they still answer correctly. Across mathematical reasoning, code generation, scientific question answering, and agentic tool-use benchmarks, we find that INT4/INT3 quantization can preserve accuracy but increase reasoning-token usage, offset"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.25519","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/2606.25519/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":"2606.25519","created_at":"2026-06-25T01:18:07.703536+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.25519v1","created_at":"2026-06-25T01:18:07.703536+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.25519","created_at":"2026-06-25T01:18:07.703536+00:00"},{"alias_kind":"pith_short_12","alias_value":"4KJIW2UUNFPB","created_at":"2026-06-25T01:18:07.703536+00:00"},{"alias_kind":"pith_short_16","alias_value":"4KJIW2UUNFPBFJVU","created_at":"2026-06-25T01:18:07.703536+00:00"},{"alias_kind":"pith_short_8","alias_value":"4KJIW2UU","created_at":"2026-06-25T01:18:07.703536+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4KJIW2UUNFPBFJVUWBL7XTIXHZ","json":"https://pith.science/pith/4KJIW2UUNFPBFJVUWBL7XTIXHZ.json","graph_json":"https://pith.science/api/pith-number/4KJIW2UUNFPBFJVUWBL7XTIXHZ/graph.json","events_json":"https://pith.science/api/pith-number/4KJIW2UUNFPBFJVUWBL7XTIXHZ/events.json","paper":"https://pith.science/paper/4KJIW2UU"},"agent_actions":{"view_html":"https://pith.science/pith/4KJIW2UUNFPBFJVUWBL7XTIXHZ","download_json":"https://pith.science/pith/4KJIW2UUNFPBFJVUWBL7XTIXHZ.json","view_paper":"https://pith.science/paper/4KJIW2UU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.25519&json=true","fetch_graph":"https://pith.science/api/pith-number/4KJIW2UUNFPBFJVUWBL7XTIXHZ/graph.json","fetch_events":"https://pith.science/api/pith-number/4KJIW2UUNFPBFJVUWBL7XTIXHZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4KJIW2UUNFPBFJVUWBL7XTIXHZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4KJIW2UUNFPBFJVUWBL7XTIXHZ/action/storage_attestation","attest_author":"https://pith.science/pith/4KJIW2UUNFPBFJVUWBL7XTIXHZ/action/author_attestation","sign_citation":"https://pith.science/pith/4KJIW2UUNFPBFJVUWBL7XTIXHZ/action/citation_signature","submit_replication":"https://pith.science/pith/4KJIW2UUNFPBFJVUWBL7XTIXHZ/action/replication_record"}},"created_at":"2026-06-25T01:18:07.703536+00:00","updated_at":"2026-06-25T01:18:07.703536+00:00"}