{"id":"https://openalex.org/W4416540656","doi":"https://doi.org/10.48550/arxiv.2504.11536","title":"ReTool: Reinforcement Learning for Strategic Tool Use in LLMs","display_name":"ReTool: Reinforcement Learning for Strategic Tool Use in LLMs","publication_year":2025,"publication_date":"2025-04-15","ids":{"openalex":"https://openalex.org/W4416540656","doi":"https://doi.org/10.48550/arxiv.2504.11536"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2504.11536","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2504.11536","pdf_url":"https://arxiv.org/pdf/2504.11536","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2504.11536","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5013569210","display_name":"Jiazhan Feng","orcid":"https://orcid.org/0000-0002-5832-6199","observed_orcids":["https://orcid.org/0000-0002-5832-6199"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feng, Jiazhan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5067715252","display_name":"Shijue Huang","orcid":"https://orcid.org/0000-0001-9443-7948","observed_orcids":["https://orcid.org/0000-0001-9443-7948"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Shijue","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101308310","display_name":"Xingwei Qu","orcid":"https://orcid.org/0000-0002-3265-7133","observed_orcids":["https://orcid.org/0000-0002-3265-7133"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qu, Xingwei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5044203241","display_name":"Ge ZHANG","orcid":"https://orcid.org/0000-0003-4694-5221","observed_orcids":["https://orcid.org/0000-0003-4694-5221","https://orcid.org/0009-0006-2000-5302"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Ge","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Qin, Yujia","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qin, Yujia","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Zhong, Baoquan","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhong, Baoquan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5021566476","display_name":"Chengquan Jiang","orcid":"https://orcid.org/0009-0004-9356-6034","observed_orcids":["https://orcid.org/0009-0004-9356-6034"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Chengquan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Chi, Jinxin","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chi, Jinxin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5019101763","display_name":"Wanjun Zhong","orcid":"https://orcid.org/0009-0007-2236-228X","observed_orcids":["https://orcid.org/0009-0007-2236-228X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhong, Wanjun","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":3,"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/T10028","display_name":"Topic Modeling","score":0.21969999372959137,"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/T10028","display_name":"Topic Modeling","score":0.21969999372959137,"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/T10906","display_name":"AI-based Problem Solving and Planning","score":0.1907999962568283,"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/T13904","display_name":"Artificial Intelligence Applications","score":0.10719999670982361,"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/code-interpreter","display_name":"code interpreter","score":0.8169999718666077},{"id":"https://openalex.org/keywords/reinforcement-learning-from-human-feedback","display_name":"reinforcement learning from human feedback","score":0.7089999914169312},{"id":"https://openalex.org/keywords/mathematical-reasoning","display_name":"mathematical reasoning","score":0.675000011920929},{"id":"https://openalex.org/keywords/aime","display_name":"AIME","score":0.5440000295639038},{"id":"https://openalex.org/keywords/neuro-symbolic-ai","display_name":"neuro-symbolic AI","score":0.6439999938011169},{"id":"https://openalex.org/keywords/deepseek-r1","display_name":"DeepSeek-R1","score":0.5109999775886536}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7398999929428101},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6446999907493591},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5920000076293945},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.5263000130653381},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.4787999987602234},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.4668999910354614},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4645000100135803},{"id":"https://openalex.org/C100776233","wikidata":"https://www.wikidata.org/wiki/Q2532492","display_name":"Bridge (graph theory)","level":2,"score":0.4465000033378601},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.4018000066280365},{"id":"https://openalex.org/C2781002164","wikidata":"https://www.wikidata.org/wiki/Q6822311","display_name":"Meta learning (computer science)","level":3,"score":0.3625999987125397},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.3540000021457672},{"id":"https://openalex.org/C77967617","wikidata":"https://www.wikidata.org/wiki/Q4677561","display_name":"Active learning (machine learning)","level":2,"score":0.31450000405311584},{"id":"https://openalex.org/C2781018962","wikidata":"https://www.wikidata.org/wiki/Q5164884","display_name":"Container (type theory)","level":2,"score":0.3116999864578247},{"id":"https://openalex.org/C115903868","wikidata":"https://www.wikidata.org/wiki/Q80993","display_name":"Software engineering","level":1,"score":0.3073999881744385},{"id":"https://openalex.org/C195344581","wikidata":"https://www.wikidata.org/wiki/Q2555318","display_name":"Automated reasoning","level":2,"score":0.30410000681877136},{"id":"https://openalex.org/C3018790387","wikidata":"https://www.wikidata.org/wiki/Q869010","display_name":"Hybrid learning","level":2,"score":0.2863999903202057},{"id":"https://openalex.org/C148220186","wikidata":"https://www.wikidata.org/wiki/Q7111912","display_name":"Outcome (game theory)","level":2,"score":0.28349998593330383},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.2833999991416931},{"id":"https://openalex.org/C4554734","wikidata":"https://www.wikidata.org/wiki/Q593744","display_name":"Knowledge base","level":2,"score":0.2831999957561493},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.2800000011920929},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.2782999873161316},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.2736999988555908},{"id":"https://openalex.org/C136643341","wikidata":"https://www.wikidata.org/wiki/Q1361526","display_name":"Reachability","level":2,"score":0.2669000029563904},{"id":"https://openalex.org/C2779439875","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Natural language understanding","level":3,"score":0.263700008392334}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2504.11536","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2504.11536","pdf_url":"https://arxiv.org/pdf/2504.11536","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2504.11536","doi":"https://doi.org/10.48550/arxiv.2504.11536","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2504.11536","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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2504.11536","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2504.11536","pdf_url":"https://arxiv.org/pdf/2504.11536","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[{"display_name":"Quality education","id":"https://metadata.un.org/sdg/4","score":0.4156000018119812}],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4416366288","https://openalex.org/W7204765869","https://openalex.org/W7117296876","https://openalex.org/W7214562811","https://openalex.org/W4415317961","https://openalex.org/W4414937509","https://openalex.org/W4416619916","https://openalex.org/W4416451863","https://openalex.org/W4417069093","https://openalex.org/W4401568174"],"abstract_inverted_index":{"While":[0],"reasoning":[1,52,70,119,233],"models":[2],"(e.g.,":[3],"DeepSeek":[4],"R1)":[5],"trained":[6],"with":[7,53,82,110,169],"reinforcement":[8],"learning":[9,92],"(RL),":[10],"excel":[11],"in":[12,17,91,181,190,211],"textual":[13],"reasoning,":[14,25],"they":[15],"struggle":[16],"scenarios":[18],"requiring":[19],"structured":[20],"problem-solving,":[21],"such":[22,203],"as":[23,131,204],"geometric":[24],"concise":[26],"computation,":[27],"or":[28],"complex":[29,231],"equation":[30],"solving-areas":[31],"where":[32],"computational":[33],"tools":[34,98],"like":[35],"code":[36,65,85,205],"interpreters":[37],"(CI)":[38],"demonstrate":[39,160],"distinct":[40],"advantages.":[41],"To":[42],"bridge":[43],"this":[44],"gap,":[45],"we":[46],"propose":[47],"ReTool,":[48],"which":[49,212],"enhances":[50],"long-form":[51,118],"tool-integrated":[54],"learning,":[55],"including":[56],"two":[57],"key":[58],"features:":[59],"(1)":[60],"dynamic":[61],"interleaving":[62],"of":[63,144,225],"real-time":[64,84],"execution":[66,86],"within":[67],"natural":[68],"language":[69],"processes,":[71],"and":[72,87,94,183,234],"(2)":[73],"an":[74,208],"automated":[75],"RL":[76,126,175],"paradigm":[77],"that":[78],"allows":[79],"policy":[80],"rollouts":[81],"multi-turn":[83],"teaches":[88],"the":[89,136,154,213,223],"model":[90,165,214],"when":[93],"how":[95],"to":[96,115,133],"invoke":[97],"based":[99],"on":[100,153],"outcome":[101],"feedback.":[102],"ReTool":[103],"employs":[104],"a":[105],"systematic":[106],"training":[107,127,171],"framework,":[108],"beginning":[109],"synthetic":[111],"cold-start":[112],"data":[113],"generation":[114],"produce":[116],"code-augmented":[117],"traces":[120],"for":[121,229],"fine-tuning":[122],"base":[123],"models.":[124],"Subsequent":[125],"leverages":[128],"task":[129],"outcomes":[130],"rewards":[132],"iteratively":[134],"refine":[135],"model's":[137],"tool":[138,146,218,227],"use":[139],"strategy,":[140],"enabling":[141],"autonomous":[142],"discovery":[143],"optimal":[145],"invocation":[147],"patterns":[148],"without":[149],"human":[150],"priors.":[151],"Experiments":[152],"challenging":[155],"MATH":[156],"Olympiad":[157],"benchmark":[158],"AIME":[159],"ReTool's":[161],"superiority:":[162],"Our":[163],"32B":[164],"achieves":[166],"67%":[167],"accuracy":[168,189],"400":[170],"steps,":[172],"outperforming":[173],"text-based":[174],"baseline":[176],"(40%":[177],"accuracy,":[178],"1080":[179],"steps)":[180],"efficiency":[182],"performance.":[184],"Remarkably,":[185],"ReTool-32B":[186],"attains":[187],"72.5%":[188],"extended":[191],"settings,":[192],"surpassing":[193],"OpenAI's":[194],"o1-preview":[195],"by":[196],"27.9%.":[197],"Further":[198],"analysis":[199],"reveals":[200],"emergent":[201],"behaviors":[202],"self-correction,":[206],"signaling":[207],"''aha":[209],"moment''":[210],"autonomously":[215],"masters":[216],"adaptive":[217],"use.":[219],"These":[220],"findings":[221],"highlight":[222],"promise":[224],"outcome-driven":[226],"integration":[228],"advancing":[230],"mathematical":[232],"offer":[235],"new":[236],"insights":[237],"into":[238],"hybrid":[239],"neuro-symbolic":[240],"systems.":[241]},"counts_by_year":[{"year":2026,"cited_by_count":2},{"year":2025,"cited_by_count":1}],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2025-10-10T00:00:00"}