{"id":"https://openalex.org/W7165942897","doi":"https://doi.org/10.48550/arxiv.2606.25519","title":"Quantization Inflates Reasoning: Token Inflation as a Hidden Cost of Low-Bit Reasoning Models","display_name":"Quantization Inflates Reasoning: Token Inflation as a Hidden Cost of Low-Bit Reasoning Models","publication_year":2026,"publication_date":"2026-06-24","ids":{"openalex":"https://openalex.org/W7165942897","doi":"https://doi.org/10.48550/arxiv.2606.25519"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.25519","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.25519","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.25519","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5063628500","display_name":"Xinyu Lian","orcid":"https://orcid.org/0009-0003-8143-4862"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lian, Xinyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5030908997","display_name":"Walid Krichene","orcid":"https://orcid.org/0000-0002-1066-9686"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Krichene, Walid","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5077623767","display_name":"Beichen Huang","orcid":"https://orcid.org/0009-0004-4951-013X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Beichen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139371842","display_name":"Masahiro Tanaka","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tanaka, Masahiro","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135375500","display_name":"Olatunji Ruwase","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ruwase, Olatunji","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139314829","display_name":"Li Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139359117","display_name":"Minjia Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Minjia","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.34599998593330383,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.34599998593330383,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.22130000591278076,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.06480000168085098,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.8804000020027161},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.8575000166893005},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.7340999841690063},{"id":"https://openalex.org/keywords/measure","display_name":"Measure (data warehouse)","score":0.48809999227523804},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.37959998846054077},{"id":"https://openalex.org/keywords/inflation","display_name":"Inflation (cosmology)","score":0.34389999508857727},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.31029999256134033}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.8804000020027161},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.8575000166893005},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7340999841690063},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6055999994277954},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.48809999227523804},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.413100004196167},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.37959998846054077},{"id":"https://openalex.org/C200941418","wikidata":"https://www.wikidata.org/wiki/Q273508","display_name":"Inflation (cosmology)","level":2,"score":0.34389999508857727},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.31029999256134033},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.29789999127388},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2838999927043915},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.2775999903678894},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.27149999141693115},{"id":"https://openalex.org/C195344581","wikidata":"https://www.wikidata.org/wiki/Q2555318","display_name":"Automated reasoning","level":2,"score":0.2687000036239624},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.2635999917984009},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.26339998841285706},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.26100000739097595},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.2547999918460846}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.25519","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.25519","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.25519","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.25519","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Quantization":[0],"is":[1,19,117],"widely":[2],"used":[3],"to":[4],"reduce":[5],"the":[6,82,92,123],"inference":[7],"cost":[8],"of":[9,49],"large":[10],"language":[11],"models,":[12],"but":[13,77],"its":[14],"effect":[15],"on":[16],"reasoning":[17,43,99,124,187],"models":[18,44,105],"not":[20],"fully":[21],"captured":[22],"by":[23,119],"final-answer":[24],"accuracy":[25,76,168,183],"or":[26],"per-token":[27,84],"latency.":[28],"We":[29,111],"show":[30,113],"that":[31,71,114,150,176],"low-bit":[32],"post-training":[33],"quantization":[34,73],"can":[35,74],"introduce":[36,91],"a":[37],"hidden":[38],"test-time":[39],"compute":[40],"cost:":[41],"quantized":[42,102,186],"often":[45],"generate":[46],"longer":[47],"chains":[48],"thought":[50],"even":[51],"when":[52,184],"they":[53],"still":[54],"answer":[55],"correctly.":[56],"Across":[57],"mathematical":[58],"reasoning,":[59],"code":[60],"generation,":[61],"scientific":[62],"question":[63],"answering,":[64],"and":[65,103,130,148,152,170],"agentic":[66],"tool-use":[67],"benchmarks,":[68],"we":[69,90,144],"find":[70,149],"INT4/INT3":[72],"preserve":[75],"increase":[78],"reasoning-token":[79,177],"usage,":[80],"offsetting":[81],"expected":[83],"speedup.":[85],"To":[86],"measure":[87],"this":[88],"effect,":[89],"CoT":[93],"Token":[94],"Inflation":[95],"Ratio,":[96],"which":[97],"compares":[98],"length":[100],"between":[101],"full-precision":[104],"averaged":[106],"across":[107],"all":[108],"evaluation":[109],"benchmarks.":[110],"further":[112],"token":[115,171],"inflation":[116],"accompanied":[118],"behavioral":[120],"changes":[121,135],"in":[122,165],"trace,":[125],"including":[126],"more":[127,163],"intermediate":[128],"steps":[129],"greater":[131],"semantic":[132],"repetition.":[133],"These":[134],"translate":[136],"into":[137],"measurable":[138],"end-to-end":[139],"real-world":[140],"serving":[141],"penalties.":[142],"Finally,":[143],"evaluate":[145],"mitigation":[146],"strategies":[147],"prompting":[151],"decoding-time":[153],"sampling":[154],"offer":[155],"inconsistent":[156],"accuracy-length":[157],"trade-offs,":[158],"while":[159],"quantization-aware":[160],"training":[161],"shows":[162],"promise":[164],"reducing":[166],"both":[167],"degradation":[169],"inflation.":[172],"Our":[173],"results":[174],"suggest":[175],"usage":[178],"should":[179],"be":[180],"reported":[181],"alongside":[182],"evaluating":[185],"models.":[188]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-26T00:00:00"}