{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:KZDVXHMN6OXBF3ESRTXYRQTNPG","short_pith_number":"pith:KZDVXHMN","schema_version":"1.0","canonical_sha256":"56475b9d8df3ae12ec928cef88c26d79b2c031c404fb081e55ef01c218e157f5","source":{"kind":"arxiv","id":"2401.03065","version":1},"attestation_state":"computed","paper":{"title":"CRUXEval: A Benchmark for Code Reasoning, Understanding and Execution","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"CRUXEval benchmark reveals GPT-4 with chain-of-thought reaches only 75% and 81% on input and output prediction for short Python functions.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.SE","authors_text":"Alex Gu, Armando Solar-Lezama, Baptiste Rozi\\`ere, Gabriel Synnaeve, Hugh Leather, Sida I. Wang","submitted_at":"2024-01-05T20:53:51Z","abstract_excerpt":"We present CRUXEval (Code Reasoning, Understanding, and eXecution Evaluation), a benchmark consisting of 800 Python functions (3-13 lines). Each function comes with an input-output pair, leading to two natural tasks: input prediction and output prediction. First, we propose a generic recipe for generating our execution benchmark which can be used to create future variation of the benchmark. Second, we evaluate twenty code models on our benchmark and discover that many recent high-scoring models on HumanEval do not show the same improvements on our benchmark. Third, we show that simple CoT and "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2401.03065","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.SE","submitted_at":"2024-01-05T20:53:51Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"1700e7185c043a55a5a1a9139892df67d8d35d960960ccb5cd1594d2edcc500b","abstract_canon_sha256":"4c190e234dc0fa17a55b121a20712b031c4c0e5b120b7649118471d954ff9ab4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:21.764527Z","signature_b64":"ORuq58uChyW2tOZBhUqpYbpxgtNmGjTMDtTyAyQSP0aFD3bTFON6W9WmNYhwJ7SpASGxcRThu4Jpk/P+Vl3sAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"56475b9d8df3ae12ec928cef88c26d79b2c031c404fb081e55ef01c218e157f5","last_reissued_at":"2026-05-17T23:39:21.763738Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:21.763738Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CRUXEval: A Benchmark for Code Reasoning, Understanding and Execution","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"CRUXEval benchmark reveals GPT-4 with chain-of-thought reaches only 75% and 81% on input and output prediction for short Python functions.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.SE","authors_text":"Alex Gu, Armando Solar-Lezama, Baptiste Rozi\\`ere, Gabriel Synnaeve, Hugh Leather, Sida I. Wang","submitted_at":"2024-01-05T20:53:51Z","abstract_excerpt":"We present CRUXEval (Code Reasoning, Understanding, and eXecution Evaluation), a benchmark consisting of 800 Python functions (3-13 lines). Each function comes with an input-output pair, leading to two natural tasks: input prediction and output prediction. First, we propose a generic recipe for generating our execution benchmark which can be used to create future variation of the benchmark. Second, we evaluate twenty code models on our benchmark and discover that many recent high-scoring models on HumanEval do not show the same improvements on our benchmark. Third, we show that simple CoT and "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The best setup, GPT-4 with chain of thought (CoT), achieves a pass@1 of 75% and 81% on input and output prediction, respectively. In contrast, Code Llama 34B achieves a pass@1 of 50% and 46% on input and output prediction.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the 800 generated functions and their input-output pairs sufficiently represent the range of code reasoning and execution challenges encountered in practical programming.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CRUXEval benchmark shows current code models including GPT-4 achieve at most 81% on input and output prediction for short Python functions, exposing gaps not captured by HumanEval.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"CRUXEval benchmark reveals GPT-4 with chain-of-thought reaches only 75% and 81% on input and output prediction for short Python functions.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"bb1b6de01df9b6b90282bff110c4970e26dc6aa03e2ecfa9841da8aba52b3735"},"source":{"id":"2401.03065","kind":"arxiv","version":1},"verdict":{"id":"09e88c37-4f05-4deb-b894-f234e1042fde","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:53:05.420039Z","strongest_claim":"The best setup, GPT-4 with chain of thought (CoT), achieves a pass@1 of 75% and 81% on input and output prediction, respectively. In contrast, Code Llama 34B achieves a pass@1 of 50% and 46% on input and output prediction.","one_line_summary":"CRUXEval benchmark shows current code models including GPT-4 achieve at most 81% on input and output prediction for short Python functions, exposing gaps not captured by HumanEval.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the 800 generated functions and their input-output pairs sufficiently represent the range of code reasoning and execution challenges encountered in practical programming.","pith_extraction_headline":"CRUXEval benchmark reveals GPT-4 with chain-of-thought reaches only 75% and 81% on input and output prediction for short Python functions."},"references":{"count":26,"sample":[{"doi":"","year":2017,"title":"A parallel corpus of Python functions and documentation strings for automated code documentation and code generation","work_id":"3484b4d3-268c-4c6a-8eb5-315f36c3de9a","ref_index":1,"cited_arxiv_id":"1707.02275","is_internal_anchor":true},{"doi":"","year":2023,"title":"Rishi Hazra, Gabriele Venturato, Pedro Zuidberg Dos Martires, and Luc De Raedt","work_id":"f145d48d-6da0-4a3a-945b-dc1551f01aff","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1909,"title":"CodeSearchNet Challenge: Evaluating the State of Semantic Code Search","work_id":"c800e7c9-4d30-4cea-a30c-2a54a64f9229","ref_index":3,"cited_arxiv_id":"1909.09436","is_internal_anchor":true},{"doi":"","year":2020,"title":"arXiv preprint arXiv:2305.05383 , year=","work_id":"613a2c7a-bd4f-49f6-b523-a324f88dbcbd","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"arXiv preprint arXiv:2401.00595 , year =","work_id":"9571587f-84df-407b-baa8-6c9557faa0c7","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":26,"snapshot_sha256":"124f6b556f9c828507a7b1f21001e1d271f7424db06bc7ee58628e01f61a3dd8","internal_anchors":2},"formal_canon":{"evidence_count":1,"snapshot_sha256":"2eb634cffbe069b3d2e2bcf1fd1717642378db482061e29b9b55f06504f0d28e"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2401.03065","created_at":"2026-05-17T23:39:21.763863+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.03065v1","created_at":"2026-05-17T23:39:21.763863+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.03065","created_at":"2026-05-17T23:39:21.763863+00:00"},{"alias_kind":"pith_short_12","alias_value":"KZDVXHMN6OXB","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"KZDVXHMN6OXBF3ES","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"KZDVXHMN","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":33,"internal_anchor_count":33,"sample":[{"citing_arxiv_id":"2402.09664","citing_title":"CodeMind: Evaluating Large Language Models for Code Reasoning","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2504.21751","citing_title":"CodeFlowBench: A Multi-turn, Iterative Benchmark for Complex Code Generation","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18851","citing_title":"STRIDE: Learnable Stepwise Language Feedback for LLM Reasoning","ref_index":50,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17958","citing_title":"Enhancing the Code Reasoning Capabilities of LLMs via Consistency-based Reinforcement Learning","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18747","citing_title":"Code as Agent Harness","ref_index":133,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18073","citing_title":"A-ProS: Towards Reliable Autonomous Programming Through Multi-Model Feedback","ref_index":76,"is_internal_anchor":true},{"citing_arxiv_id":"2507.14913","citing_title":"PromptSuite: A Task-Agnostic Framework for Multi-Prompt Generation","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2510.15494","citing_title":"Do AI Models Dream of Faster Code? An Empirical Study on LLM-Proposed Performance Improvements in Real-World Software","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2510.15079","citing_title":"Assessing Coherency and Consistency of Code Execution Reasoning by Large Language Models","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2512.00127","citing_title":"Generating Verifiable Chain of Thoughts from Exection-Traces","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2512.14917","citing_title":"Evaluating Code Reasoning Abilities of Large Language Models Under Real-World Settings","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2601.15593","citing_title":"Parallelism and Generation Order in Masked Diffusion Language Models: Limits Today, Potential Tomorrow","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2603.14628","citing_title":"s2n-bignum-bench: A practical benchmark for evaluating low-level code reasoning of LLMs","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2512.15745","citing_title":"LLaDA2.0: Scaling Up Diffusion Language Models to 100B","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2406.00515","citing_title":"A Survey on Large Language Models for Code Generation","ref_index":82,"is_internal_anchor":true},{"citing_arxiv_id":"2604.03144","citing_title":"InCoder-32B-Thinking: Industrial Code World Model for Thinking","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11922","citing_title":"StepCodeReasoner: Aligning Code Reasoning with Stepwise Execution Traces via Reinforcement Learning","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2402.19173","citing_title":"StarCoder 2 and The Stack v2: The Next Generation","ref_index":208,"is_internal_anchor":true},{"citing_arxiv_id":"2604.28139","citing_title":"Claw-Eval-Live: A Live Agent Benchmark for Evolving Real-World Workflows","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2604.25399","citing_title":"CoRE: A Fine-Grained Code Reasoning Benchmark Beyond Output Prediction","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06184","citing_title":"Teaching LLMs Program Semantics via Symbolic Execution Traces","ref_index":19,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06219","citing_title":"Joint Consistency: A Unified Test-Time Aggregation Framework via Energy Minimization","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2604.21505","citing_title":"Assessing the Impact of Requirement Ambiguity on LLM-based Function-Level Code Generation","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07353","citing_title":"Confidence-Aware Alignment Makes Reasoning LLMs More Reliable","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2604.05560","citing_title":"An Iterative Test-and-Repair Framework for Competitive Code Generation","ref_index":14,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":1,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KZDVXHMN6OXBF3ESRTXYRQTNPG","json":"https://pith.science/pith/KZDVXHMN6OXBF3ESRTXYRQTNPG.json","graph_json":"https://pith.science/api/pith-number/KZDVXHMN6OXBF3ESRTXYRQTNPG/graph.json","events_json":"https://pith.science/api/pith-number/KZDVXHMN6OXBF3ESRTXYRQTNPG/events.json","paper":"https://pith.science/paper/KZDVXHMN"},"agent_actions":{"view_html":"https://pith.science/pith/KZDVXHMN6OXBF3ESRTXYRQTNPG","download_json":"https://pith.science/pith/KZDVXHMN6OXBF3ESRTXYRQTNPG.json","view_paper":"https://pith.science/paper/KZDVXHMN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.03065&json=true","fetch_graph":"https://pith.science/api/pith-number/KZDVXHMN6OXBF3ESRTXYRQTNPG/graph.json","fetch_events":"https://pith.science/api/pith-number/KZDVXHMN6OXBF3ESRTXYRQTNPG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KZDVXHMN6OXBF3ESRTXYRQTNPG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KZDVXHMN6OXBF3ESRTXYRQTNPG/action/storage_attestation","attest_author":"https://pith.science/pith/KZDVXHMN6OXBF3ESRTXYRQTNPG/action/author_attestation","sign_citation":"https://pith.science/pith/KZDVXHMN6OXBF3ESRTXYRQTNPG/action/citation_signature","submit_replication":"https://pith.science/pith/KZDVXHMN6OXBF3ESRTXYRQTNPG/action/replication_record"}},"created_at":"2026-05-17T23:39:21.763863+00:00","updated_at":"2026-05-17T23:39:21.763863+00:00"}