{"paper":{"title":"RepoBench: Benchmarking Repository-Level Code Auto-Completion Systems","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"RepoBench introduces a benchmark for repository-level code auto-completion with three tasks covering retrieval, next-line prediction, and combined pipelines in Python and Java.","cross_cats":["cs.AI","cs.SE"],"primary_cat":"cs.CL","authors_text":"Canwen Xu, Julian McAuley, Tianyang Liu","submitted_at":"2023-06-05T17:59:41Z","abstract_excerpt":"Large Language Models (LLMs) have greatly advanced code auto-completion systems, with a potential for substantial productivity enhancements for developers. However, current benchmarks mainly focus on single-file tasks, leaving an assessment gap for more complex, real-world, multi-file programming scenarios. To fill this gap, we introduce RepoBench, a new benchmark specifically designed for evaluating repository-level code auto-completion systems. RepoBench supports both Python and Java and consists of three interconnected evaluation tasks: RepoBench-R (Retrieval), RepoBench-C (Code Completion)"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Current benchmarks mainly focus on single-file tasks, leaving an assessment gap for more complex, real-world, multi-file programming scenarios. To fill this gap, we introduce RepoBench.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the three constructed tasks and data selection in RepoBench faithfully capture the challenges of real repository-level code completion without introducing selection biases or artificial simplifications.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"RepoBench is a new benchmark with retrieval, completion, and pipeline tasks to evaluate code auto-completion systems on entire repositories instead of single files.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"RepoBench introduces a benchmark for repository-level code auto-completion with three tasks covering retrieval, next-line prediction, and combined pipelines in Python and Java.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e972ce66cd53486dc9705e68eb3fe3a7309e02a2d3b020bf001d0a4ef0d4432b"},"source":{"id":"2306.03091","kind":"arxiv","version":2},"verdict":{"id":"3dadf7f7-8731-4710-8d51-daec6e67906c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T22:26:46.801004Z","strongest_claim":"Current benchmarks mainly focus on single-file tasks, leaving an assessment gap for more complex, real-world, multi-file programming scenarios. To fill this gap, we introduce RepoBench.","one_line_summary":"RepoBench is a new benchmark with retrieval, completion, and pipeline tasks to evaluate code auto-completion systems on entire repositories instead of single files.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the three constructed tasks and data selection in RepoBench faithfully capture the challenges of real repository-level code completion without introducing selection biases or artificial simplifications.","pith_extraction_headline":"RepoBench introduces a benchmark for repository-level code auto-completion with three tasks covering retrieval, next-line prediction, and combined pipelines in Python and Java."},"references":{"count":55,"sample":[{"doi":"","year":2023,"title":"Colt5: Faster long-range transformers with conditional computation, 2023","work_id":"78a986d9-408b-420e-a50a-7edc132af650","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Santacoder: don’t reach for the stars!,","work_id":"5164e1be-7a21-42aa-a4b1-ef335098c626","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Santacoder: don’t reach for the stars! arXiv preprint arXiv:2301.03988","work_id":"bf393c50-a11b-4a0f-8513-52428ede71f7","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1109/msr.2013.6624029","year":2013,"title":"In: 2013 10th Working Conference on Mining Software Repositories (MSR), pp 207--216, doi:10.1109/MSR.2013.6624029","work_id":"5d216e8e-7c15-47d4-b59e-ede90680ea91","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","ref_index":5,"cited_arxiv_id":"2108.07732","is_internal_anchor":true}],"resolved_work":55,"snapshot_sha256":"0f2dc887476a6b7da5163cb2f911da85c091d02251273c1c39cd21ee1bce65e4","internal_anchors":8},"formal_canon":{"evidence_count":1,"snapshot_sha256":"fed04b1cdee6e4ad3748e16db268020938930d0b225749e03aae499d72e323e3"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}