{"id":"https://openalex.org/W7152558338","doi":"https://doi.org/10.48550/arxiv.2604.06376","title":"MTA-Agent: An Open Recipe for Multimodal Deep Search Agents","display_name":"MTA-Agent: An Open Recipe for Multimodal Deep Search Agents","publication_year":2026,"publication_date":"2026-04-07","ids":{"openalex":"https://openalex.org/W7152558338","doi":"https://doi.org/10.48550/arxiv.2604.06376"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.06376","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.06376","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":null,"license_id":null,"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.2604.06376","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5133259803","display_name":"Xiangyu Peng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peng, Xiangyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133257799","display_name":"Can Qin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qin, Can","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076614285","display_name":"An Yan","orcid":"https://orcid.org/0009-0009-2778-3803"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yan, An","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133309128","display_name":"Xinyi Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Xinyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133263717","display_name":"Zeyuan Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Zeyuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133264964","display_name":"Ran Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Ran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5066791810","display_name":"Chien-Sheng Wu","orcid":"https://orcid.org/0000-0002-5598-5324"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Chien-Sheng","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":0.9750000238418579,"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":0.9750000238418579,"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/T10028","display_name":"Topic Modeling","score":0.00860000029206276,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.002099999925121665,"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/pipeline","display_name":"Pipeline (software)","score":0.6330999732017517},{"id":"https://openalex.org/keywords/consistency","display_name":"Consistency (knowledge bases)","score":0.6157000064849854},{"id":"https://openalex.org/keywords/recipe","display_name":"Recipe","score":0.5946000218391418},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.5860999822616577},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.40630000829696655},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.3659999966621399}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7932999730110168},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.6330999732017517},{"id":"https://openalex.org/C2776436953","wikidata":"https://www.wikidata.org/wiki/Q5163215","display_name":"Consistency (knowledge bases)","level":2,"score":0.6157000064849854},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6068999767303467},{"id":"https://openalex.org/C2778671685","wikidata":"https://www.wikidata.org/wiki/Q219239","display_name":"Recipe","level":2,"score":0.5946000218391418},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.5860999822616577},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4348999857902527},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.40630000829696655},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3659999966621399},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.3449000120162964},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.32359999418258667},{"id":"https://openalex.org/C36464697","wikidata":"https://www.wikidata.org/wiki/Q451553","display_name":"Visualization","level":2,"score":0.314300000667572},{"id":"https://openalex.org/C141917322","wikidata":"https://www.wikidata.org/wiki/Q1025017","display_name":"Cache coherence","level":5,"score":0.26660001277923584}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.06376","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.06376","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.06376","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.06376","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"score":0.48678362369537354,"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Multimodal":[0],"large":[1],"language":[2],"models":[3],"(MLLMs)":[4],"have":[5],"demonstrated":[6],"strong":[7],"capabilities":[8],"in":[9,16],"visual":[10,26,75],"understanding,":[11],"yet":[12],"they":[13],"remain":[14],"limited":[15],"complex,":[17],"multi-step":[18],"reasoning":[19,163],"that":[20,156,190],"requires":[21],"deep":[22,218],"searching":[23],"and":[24,65,70,76,79,117,145,165,178,183,227,233],"integrating":[25],"evidence":[27,72],"with":[28],"external":[29],"knowledge.":[30],"In":[31],"this":[32,36],"work,":[33],"we":[34,188,208,220],"address":[35],"challenge":[37],"by":[38,199],"constructing":[39],"high-quality,":[40],"verified":[41],"multi-hop":[42,82,102],"vision-language":[43],"training":[44,96,157,191,205,225],"data":[45,105,160],"for":[46,56,216],"multimodal":[47,125,217,238],"deep-search":[48],"agents.":[49,240],"We":[50,153],"propose":[51],"a":[52,94,109,122,212],"Multi-hop":[53],"Tool-Augmented":[54],"Agent":[55],"Evidence-based":[57],"QA":[58],"Synthesis":[59],"(MTA-Agent),":[60],"which":[61],"automatically":[62],"selects":[63],"tools":[64],"their":[66],"parameters":[67],"to":[68,113,176,180,230],"retrieve":[69],"validate":[71],"from":[73,86,174],"both":[74,162],"textual":[77],"sources":[78],"generates":[80],"structured":[81],"question-answer":[83],"trajectories.":[84],"Starting":[85],"diverse":[87],"VQA":[88],"seed":[89],"datasets,":[90],"our":[91,159],"pipeline":[92],"produces":[93],"large-scale":[95],"dataset,":[97,224],"MTA-Vision-DeepSearch,":[98,121],"containing":[99],"21K":[100],"high-quality":[101],"examples.":[103],"The":[104],"is":[106],"filtered":[107],"through":[108],"multi-stage":[110],"verification":[111],"process":[112],"ensure":[114],"factual":[115],"consistency":[116],"answer":[118],"uniqueness.":[119],"Using":[120],"32B":[123],"open-source":[124],"search":[126,185,239],"agent":[127],"achieves":[128],"state-of-the-art":[129],"performance,":[130],"reaching":[131],"an":[132],"average":[133,170],"of":[134,172],"54.63\\%":[135],"across":[136],"six":[137],"challenging":[138],"benchmarks,":[139],"outperforming":[140],"GPT-5":[141],"(51.86\\%),":[142],"Gemini-2.5-Pro":[143],"(50.98\\%),":[144],"Gemini-3-Pro":[146],"(54.46\\%)":[147],"under":[148],"the":[149,169,222],"same":[150],"tool":[151,197],"settings.":[152],"further":[154],"show":[155],"on":[158,236],"improves":[161],"depth":[164],"tool-use":[166],"behavior,":[167],"increasing":[168],"number":[171],"steps":[173],"2.27":[175],"4.28,":[177],"leading":[179],"more":[181],"systematic":[182],"persistent":[184],"strategies.":[186],"Additionally,":[187],"demonstrate":[189],"can":[192],"be":[193],"performed":[194],"without":[195],"real-time":[196],"calls":[198],"replaying":[200],"cached":[201],"interactions,":[202],"significantly":[203],"reducing":[204],"cost.":[206],"Importantly,":[207],"present":[209],"MTA-Agent":[210],"as":[211],"fully":[213],"open":[214,237],"recipe":[215],"search:":[219],"release":[221],"entire":[223],"trajectories,":[226],"implementation":[228],"details":[229],"enable":[231],"reproducibility":[232],"future":[234],"research":[235]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-10T00:00:00"}