{"id":"https://openalex.org/W7154375159","doi":"https://doi.org/10.48550/arxiv.2604.11465","title":"Three Roles, One Model: Role Orchestration at Inference Time to Close the Performance Gap Between Small and Large Agents","display_name":"Three Roles, One Model: Role Orchestration at Inference Time to Close the Performance Gap Between Small and Large Agents","publication_year":2026,"publication_date":"2026-04-13","ids":{"openalex":"https://openalex.org/W7154375159","doi":"https://doi.org/10.48550/arxiv.2604.11465"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.11465","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.11465","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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.2604.11465","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5133579314","display_name":"S. Aaron McClendon","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"McClendon, S. Aaron","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133554208","display_name":"Jorge Gallego-Feliciano","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gallego-Feliciano, Jorge","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5085156750","display_name":"Stavros Zervoudakis","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zervoudakis, Stavros","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5133558049","display_name":"Antonios Saravanos","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Saravanos, Antonios","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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.12309999763965607,"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"}},"topics":[{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.12309999763965607,"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"}},{"id":"https://openalex.org/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.10159999877214432,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.08020000159740448,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/automatic-summarization","display_name":"Automatic summarization","score":0.6104000210762024},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.5889999866485596},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5839999914169312},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.5611000061035156},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.5523999929428101},{"id":"https://openalex.org/keywords/probabilistic-logic","display_name":"Probabilistic logic","score":0.4205999970436096},{"id":"https://openalex.org/keywords/orchestration","display_name":"Orchestration","score":0.3813000023365021},{"id":"https://openalex.org/keywords/conversation","display_name":"Conversation","score":0.3774999976158142}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7960000038146973},{"id":"https://openalex.org/C170858558","wikidata":"https://www.wikidata.org/wiki/Q1394144","display_name":"Automatic summarization","level":2,"score":0.6104000210762024},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.5889999866485596},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5839999914169312},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5611000061035156},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.5523999929428101},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4271000027656555},{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.4205999970436096},{"id":"https://openalex.org/C199168358","wikidata":"https://www.wikidata.org/wiki/Q3367000","display_name":"Orchestration","level":3,"score":0.3813000023365021},{"id":"https://openalex.org/C2777200299","wikidata":"https://www.wikidata.org/wiki/Q52943","display_name":"Conversation","level":2,"score":0.3774999976158142},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.3716999888420105},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.36559998989105225},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.3425000011920929},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3125},{"id":"https://openalex.org/C2780598303","wikidata":"https://www.wikidata.org/wiki/Q65921492","display_name":"Flexibility (engineering)","level":2,"score":0.3050000071525574},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.29670000076293945},{"id":"https://openalex.org/C158154518","wikidata":"https://www.wikidata.org/wiki/Q7310970","display_name":"Relevance (law)","level":2,"score":0.29429998993873596},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.2851000130176544},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.2736000120639801},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.26170000433921814},{"id":"https://openalex.org/C134652429","wikidata":"https://www.wikidata.org/wiki/Q1052698","display_name":"Jitter","level":2,"score":0.26019999384880066}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.11465","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.11465","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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.2604.11465","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.11465","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"model":[2,38,68,98,106,123,135,197],"(LLM)":[3],"agents":[4,14],"show":[5],"promise":[6],"on":[7,15,44,52,180],"realistic":[8],"tool-use":[9],"tasks,":[10],"but":[11],"deploying":[12],"capable":[13],"modest":[16],"hardware":[17],"remains":[18],"challenging.":[19],"We":[20,224],"study":[21],"whether":[22],"inference-time":[23,211],"scaffolding":[24,91,160],"alone,":[25],"without":[26,144],"any":[27,64],"additional":[28],"training":[29],"compute,":[30],"can":[31,213],"improve":[32],"the":[33,53,66,95,120,127,140,155,204,226,240],"performance":[34,172],"of":[35,239],"a":[36,45,81,88,104,229,233],"small":[37,215],"in":[39,99,173,255],"complex":[40],"multi-step":[41],"environments.":[42],"Operating":[43],"single":[46],"24GB":[47],"GPU,":[48],"we":[49,86],"evaluate":[50],"Qwen3-8B":[51],"AppWorld":[54,206],"benchmark":[55],"under":[56],"both":[57,174],"full-precision":[58,192],"and":[59,73,130,138,164,252],"4-bit":[60],"quantized":[61],"configurations.":[62],"Without":[63],"intervention,":[65],"raw":[67],"achieves":[69],"just":[70],"5.4%":[71],"(FP16)":[72,163],"3.0%":[74],"(AWQ)":[75,166],"task":[76,167],"goal":[77,168],"completion.":[78],"Guided":[79],"by":[80],"systematic":[82],"failure":[83,151],"mode":[84],"analysis,":[85],"introduce":[87],"three-tier":[89],"inference":[90],"pipeline":[92],"that":[93,107,124,136,209],"deploys":[94],"same":[96,156,241],"frozen":[97,234],"three":[100,237],"distinct":[101],"roles:":[102],"(1)":[103],"summarization":[105],"preserves":[108],"critical":[109],"artifacts":[110],"(tokens,":[111],"credentials,":[112],"API":[113],"responses)":[114],"while":[115],"compressing":[116],"dialogue":[117],"history;":[118],"(2)":[119],"main":[121],"agent":[122],"reasons":[125],"over":[126,232],"compressed":[128],"context;":[129],"(3)":[131],"an":[132],"isolated":[133],"correction":[134],"reviews":[137],"revises":[139],"agent's":[141],"code":[142],"output":[143],"access":[145],"to":[146,154,184,188,248],"conversation":[147],"history,":[148],"breaking":[149],"repetitive":[150],"loops.":[152],"Applied":[153],"unmodified":[157],"model,":[158,236],"this":[159],"yields":[161],"8.9%":[162],"5.9%":[165],"completion,":[169],"roughly":[170],"doubling":[171],"settings,":[175],"with":[176,218,243],"particularly":[177],"strong":[178],"gains":[179],"difficulty-1":[181],"tasks":[182],"(15.8%":[183],"26.3%":[185],"FP16;":[186],"5.3%":[187],"14.0%":[189],"AWQ).":[190],"On":[191],"inference,":[193],"our":[194],"scaffolded":[195,230],"8B":[196],"surpasses":[198],"DeepSeek-Coder":[199],"33B":[200],"Instruct":[201],"(7.1%)":[202],"from":[203],"original":[205],"evaluation,":[207],"demonstrating":[208],"structured":[210],"interventions":[212],"make":[214],"models":[216],"competitive":[217],"systems":[219],"4":[220],"times":[221],"their":[222],"size.":[223],"formalize":[225],"approach":[227],"as":[228],"policy":[231],"base":[235],"invocations":[238],"weights":[242],"different":[244],"conditioning,":[245],"drawing":[246],"connections":[247],"test-time":[249],"compute":[250],"scaling":[251],"action-space":[253],"shaping":[254],"reinforcement":[256],"learning.":[257]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-04-15T00:00:00"}