CoChat
Sign Up
Cited by 420Software Engineering ResearchSoftware Testing and Debugging TechniquesTeaching and Learning Programming

Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation

Open access
Liu, Jiawei, Xia, Chunqiu Steven, Wang, Yuyao, Zhang, Lingming·2023·Advances in Neural Information Processing Systems 36
DOI 10.52202/075280-0943
Open this paper in CoChat — free accountOpen PDF

Abstract

Program synthesis has been long studied with recent approaches focused on directly using the power of Large Language Models (LLMs) to generate code. Programming benchmarks, with curated synthesis problems and test-cases, are used to measure the performance of various LLMs on code synthesis. However, these test-cases can be limited in both quantity and quality for fully assessing the functional correctness of the generated code. Such limitation in the existing benchmarks begs the following question: In the era of LLMs, is the code generated really correct? To answer this, we propose EvalPlus -- a code synthesis evaluation framework to rigorously benchmark the functional correctness of LLM-synthesized code. EvalPlus augments a given evaluation dataset with large amounts of test-cases newly produced by an automatic test input generator, powered by both LLM- and mutation-based strategies. While EvalPlus is general, we extend the test-cases of the popular HumanEval benchmark by 80x to build HumanEval+. Our extensive evaluation across 26 popular LLMs (e.g., GPT-4 and ChatGPT) demonstrates that HumanEval+ is able to catch significant amounts of previously undetected wrong code synthesized by LLMs, reducing the pass@k by up-to 19.3-28.9%. We also surprisingly found that test insufficiency can lead to mis-ranking. For example, both WizardCoder-CodeLlama and Phind-CodeLlama now outperform ChatGPT on HumanEval+, while none of them could on HumanEval. Our work not only indicates that prior popular code synthesis evaluation results do not accurately reflect the true performance of LLMs for code synthesis, but also opens up a new direction to improve such programming benchmarks through automated testing. We have open-sourced our tools, enhanced datasets as well as all LLM-generated code at https://github.com/evalplus/evalplus to facilitate and accelerate future LLM-for-code research.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.52202/075280-0943

Cites

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • MBPP+ offers 35 times more test cases than the original MBPP.

    “Similarly, MBPP+ offers 35 times more test cases than the original MBPP.”

    From Qwen2.5-Coder Technical Report (2024) · cites this paper as Liu et al., 2023

    The Conclusion & Future Work section says the authors plan to apply EvalPlus to MBPP in the future, rather than reporting an MBPP+ benchmark. The paper reports an 80-fold test-case increase for HumanEval+, not a 35-fold increase for MBPP+. (LLM, conf=0.86)

    Checked by a reasoning model — may be wrong.

    Experimental — accuracy not yet measured

  • A finite test suite necessarily under-approximates an executable's full specification.

    “For the second case, a finite test suite necessarily under-approximates an executable's full specification (Liu et al., 2023; Le Goues et al., 2015).”

    From ProgramBench: Can Language Models Rebuild Programs From Scratch? (2026) · cites this paper as Liu et al., 2023

    The Introduction (page not available in our copy) says benchmark tests often fail to capture all possible scenarios and that flawed code can pass them. That supports the concern about incomplete testing, but not the claim that every finite test suite necessarily under-approximates an executable's full specification. (LLM, conf=0.86)

    Experimental — accuracy not yet measured

  • As a result of context exceeding window size, LLMs frequently produce code that fails to compile, violates existing patterns, or ignores critical constraints.

    “As a result, LLMs frequently produce code that fails to compile, violates existing patterns, or ignores critical constraints when the relevant context exceeds their window size [22].”

    From LongCodeZip: Compress Long Context for Code Language Models (2025) · cites this paper as [22]

    The Introduction says LLM-generated code can be incorrect, and §2.3 says vague descriptions can lead models to misinterpret desired functionality or input requirements. But the paper does not attribute failures to relevant context exceeding a model’s window, or establish the claimed compilation and existing-pattern failures under that condition. (LLM, conf=0.80)

    Experimental — accuracy not yet measured

  • Researchers are developing synthetic challenging benchmarks, often involving models in the construction of evaluation instances, such as subsampling instances from existing benchmarks that cannot be solved by current models or augmenting them to construct more challenging evaluations.

    “For example, some subsample instances from existing benchmarks that cannot be solved by current models [93, 82], or augment them to construct more challenging evaluations [21, 44, 49].”

    From SciCode: A Research Coding Benchmark Curated by Scientists (2024) · cites this paper as [49]

  • EvalPlus extends HumanEval into HumanEval+ by adding 80 times more unique test cases and correcting inaccurate groundtruth solutions.

    “To further ensure accurate evaluation, EvalPlus (Liu et al., 2023) extends HumanEval into HumanEval+ by adding 80 times more unique test cases and correcting inaccurate groundtruth solutions in HumanEval.”

    From Qwen2.5-Coder Technical Report (2024) · cites this paper as Liu et al., 2023

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Still being checked.

Research works printed but not linked: 62

  1. T. Ahmed and P. Devanbu. Few-shot training llms for project-specific code-summarization. In 37th IEEE/ACM International Conference on Automated Software Engineering, pages 1-5, 2022.
  2. L. B. Allal, R. Li, D. Kocetkov, C. Mou, C. Akiki, C. M. Ferrandis, N. Muennighoff, M. Mishra, A. Gu, M. Dey, et al. Santacoder: don't reach for the stars! arXiv preprint arXiv:2301.03988, 2023. · arXiv
  3. S. Bang, S. Nam, I. Chun, H. Y. Jhoo, and J. Lee. Smt-based translation validation for machine learning compiler. In Computer Aided Verification: 34th International Conference, CAV 2022, Haifa, Israel, August 7-10, 2022, Proceedings, Part II, pages 386-407. Springer, 2022.
  4. T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, et al. Language models are few-shot learners. Advances in neural information processing systems, 33:1877-1901, 2020.
  5. C. Cadar, D. Dunbar, D. R. Engler, et al. Klee: unassisted and automatic generation of high-coverage tests for complex systems programs. In OSDI, volume 8, pages 209-224, 2008.
  6. F. Cassano, J. Gouwar, D. Nguyen, S. Nguyen, L. Phipps-Costin, D. Pinckney, M.-H. Yee, Y. Zi, C. J. Anderson, M. Q. Feldman, et al. Multipl-e: A scalable and polyglot approach to benchmarking neural code generation. IEEE Transactions on Software Engineering, 2023.
  7. S. K. Cha, M. Woo, and D. Brumley. Program-adaptive mutational fuzzing. In 2015 IEEE Symposium on Security and Privacy, pages 725-741. IEEE, 2015.
  8. M. Chen, J. Tworek, H. Jun, Q. Yuan, H. P. d. O. Pinto, J. Kaplan, H. Edwards, Y. Burda, N. Joseph, G. Brockman, et al. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374, 2021. · arXiv
  9. Y. Deng, C. S. Xia, H. Peng, C. Yang, and L. Zhang. Large language models are zero-shot fuzzers: Fuzzing deep-learning libraries via large language models. In 32nd International Symposium on Software Testing and Analysis (ISSTA), 2023.
  10. Y. Deng, C. S. Xia, C. Yang, S. D. Zhang, S. Yang, and L. Zhang. Large language models are edge-case fuzzers: Testing deep learning libraries via fuzzgpt. In 46th International Conference on Software Engineering (ICSE), 2024.
  11. U. Feige. A threshold of ln n for approximating set cover. Journal of the ACM (JACM), 45(4):634-652, 1998.
  12. D. Fried, A. Aghajanyan, J. Lin, S. Wang, E. Wallace, F. Shi, R. Zhong, S. Yih, L. Zettlemoyer, and M. Lewis. Incoder: A generative model for code infilling and synthesis. In The Eleventh International Conference on Learning Representations, 2023.
  13. C. Green. Application of theorem proving to problem solving. In Readings in Artificial Intelligence, pages 202-222. Elsevier, 1981.
  14. S. Gulwani. Automating string processing in spreadsheets using input-output examples. SIGPLAN Not., 46(1):317-330, jan 2011.
  15. S. Gulwani, O. Polozov, and R. Singh. Program synthesis. Foundations and Trends® in Programming Languages, 4(1-2):1-119, 2017.
  16. T. Hastie, R. Tibshirani, J. H. Friedman, and J. H. Friedman. The elements of statistical learning: data mining, inference, and prediction, volume 2. Springer, 2009.
  17. C. Holler, K. Herzig, and A. Zeller. Fuzzing with code fragments. In 21st USENIX Security Sym- posium (USENIX Security 12), pages 445-458, Bellevue, WA, Aug. 2012. USENIX Association.
  18. M. Ivanković, G. Petrović, R. Just, and G. Fraser. Code coverage at google. In Proceedings of the 2019 27th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering, pages 955-963, 2019.
  19. S. Iyer, I. Konstas, A. Cheung, and L. Zettlemoyer. Mapping language to code in programmatic context. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1643-1652, 2018.
  20. A. Q. Jiang, A. Sablayrolles, A. Mensch, C. Bamford, D. S. Chaplot, D. d. l. Casas, F. Bressand, G. Lengyel, G. Lample, L. Saulnier, et al. Mistral 7b. arXiv preprint arXiv:2310.06825, 2023. · arXiv
  21. N. Jiang, K. Liu, T. Lutellier, and L. Tan. Impact of code language models on automated program repair. arXiv preprint arXiv:2302.05020, 2023. · arXiv
  22. C. E. Jimenez, J. Yang, A. Wettig, S. Yao, K. Pei, O. Press, and K. Narasimhan. Swe-bench: Can language models resolve real-world github issues? arXiv preprint arXiv:2310.06770, 2023. · arXiv
  23. A. Kalyan, A. Mohta, O. Polozov, D. Batra, P. Jain, and S. Gulwani. Neural-guided deductive search for real-time program synthesis from examples. In International Conference on Learning Representations, 2018.
  24. J. C. King. Symbolic execution and program testing. Communications of the ACM, 19(7):385-394, 1976.
  25. M.-A. Lachaux, B. Roziere, L. Chanussot, and G. Lample. Unsupervised translation of programming languages. arXiv preprint arXiv:2006.03511, 2020. · arXiv
  26. K. R. M. Leino. Dafny: An automatic program verifier for functional correctness. In International conference on logic for programming artificial intelligence and reasoning, pages 348-370. Springer, 2010.
  27. Y. Li, D. Choi, J. Chung, N. Kushman, J. Schrittwieser, R. Leblond, T. Eccles, J. Keeling, F. Gimeno, A. Dal Lago, et al. Competition-level code generation with alphacode. Science, 378(6624):1092-1097, 2022.
  28. J. Liu, Y. Wei, S. Yang, Y. Deng, and L. Zhang. Coverage-guided tensor compiler fuzzing with joint ir-pass mutation. Proceedings of the ACM on Programming Languages, 6(OOPSLA1):1-26, Apr. 2022.
  29. Z. Liu, C. Chen, J. Wang, X. Che, Y. Huang, J. Hu, and Q. Wang. Fill in the blank: Context-aware automated text input generation for mobile gui testing. In 2023 IEEE/ACM 45th International Conference on Software Engineering (ICSE), pages 1355-1367. IEEE, 2023.
  30. N. P. Lopes, J. Lee, C.-K. Hur, Z. Liu, and J. Regehr. Alive2: bounded translation validation for llvm. In Proceedings of the 42nd ACM SIGPLAN International Conference on Programming Language Design and Implementation, pages 65-79, 2021.
  31. S. Lu, D. Guo, S. Ren, J. Huang, A. Svyatkovskiy, A. Blanco, C. Clement, D. Drain, D. Jiang, D. Tang, et al. Codexglue: A machine learning benchmark dataset for code understanding and generation. arXiv preprint arXiv:2102.04664, 2021. · arXiv
  32. Z. Luo, C. Xu, P. Zhao, Q. Sun, X. Geng, W. Hu, C. Tao, J. Ma, Q. Lin, and D. Jiang. Wizardcoder: Empowering code large language models with evol-instruct. arXiv preprint arXiv:2306.08568, 2023. · arXiv
  33. Z. Manna and R. J. Waldinger. Toward automatic program synthesis. Communications of the ACM, 14(3):151-165, 1971.
  34. W. M. McKeeman. Differential testing for software. Digital Technical Journal, 10(1):100-107, 1998.
  35. B. Meyer. Applying'design by contract'. Computer, 25(10):40-51, 1992.
  36. B. P. Miller, L. Fredriksen, and B. So. An empirical study of the reliability of unix utilities. Communications of the ACM, 33(12):32-44, 1990.
  37. G. C. Necula. Translation validation for an optimizing compiler. In Proceedings of the ACM SIGPLAN 2000 conference on Programming language design and implementation, pages 83-94, 2000.
  38. E. Nijkamp, H. Hayashi, C. Xiong, S. Savarese, and Y. Zhou. Codegen2: Lessons for training llms on programming and natural languages. arXiv preprint, 2023.
  39. E. Nijkamp, B. Pang, H. Hayashi, L. Tu, H. Wang, Y. Zhou, S. Savarese, and C. Xiong. Codegen: An open large language model for code with multi-turn program synthesis. In The Eleventh International Conference on Learning Representations, 2023.
  40. P. Oehlert. Violating assumptions with fuzzing. IEEE Security & Privacy, 3(2):58-62, 2005.
  41. OpenAI. Gpt-4 technical report. ArXiv, abs/2303.08774, 2023. · arXiv
  42. K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311-318, 2002.
  43. G. Petrović and M. Ivanković. State of mutation testing at google. In Proceedings of the 40th international conference on software engineering: Software engineering in practice, pages 163-171, 2018.
  44. G. Rothermel, M. J. Harrold, J. Von Ronne, and C. Hong. Empirical studies of test-suite reduction. Software Testing, Verification and Reliability, 12(4):219-249, 2002.
  45. B. Rozière, J. Gehring, F. Gloeckle, S. Sootla, I. Gat, X. E. Tan, Y. Adi, J. Liu, T. Remez, J. Rapin, et al. Code llama: Open foundation models for code. arXiv preprint arXiv:2308.12950, 2023. · arXiv
  46. B. Roziere, J. M. Zhang, F. Charton, M. Harman, G. Synnaeve, and G. Lample. Leveraging automated unit tests for unsupervised code translation. arXiv preprint arXiv:2110.06773, 2021. · arXiv
  47. K. Serebryany. Continuous fuzzing with libfuzzer and addresssanitizer. In 2016 IEEE Cybersecurity Development (SecDev), pages 157-157. IEEE, 2016.
  48. D. E. Shaw, W. R. Swartout, and C. C. Green. Inferring lisp programs from examples. In IJCAI, volume 75, pages 260-267, 1975.
  49. A. Shi, A. Gyori, M. Gligoric, A. Zaytsev, and D. Marinov. Balancing trade-offs in test-suite reduction. In Proceedings of the 22nd ACM SIGSOFT international symposium on foundations of software engineering, pages 246-256, 2014.
  50. A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polo- sukhin. Attention is all you need. Advances in neural information processing systems, 30, 2017.
  51. R. J. Waldinger and R. C. Lee. Prow: A step toward automatic program writing. In Proceedings of the 1st international joint conference on Artificial intelligence, pages 241-252, 1969.
  52. Y. Wang, H. Le, A. D. Gotmare, N. D. Bui, J. Li, and S. C. Hoi. Codet5+: Open code large language models for code understanding and generation. arXiv preprint arXiv:2305.07922, 2023. · arXiv
  53. Y. Wei, C. S. Xia, and L. Zhang. Copiloting the copilots: Fusing large language models with completion engines for automated program repair. arXiv preprint arXiv:2309.00608, 2023. · arXiv
  54. D. Winterer, C. Zhang, and Z. Su. On the unusual effectiveness of type-aware operator mutations for testing smt solvers. Proceedings of the ACM on Programming Languages, 4(OOPSLA):1-25, 2020.
  55. C. S. Xia, M. Paltenghi, J. L. Tian, M. Pradel, and L. Zhang. Universal fuzzing via large language models. In 46th International Conference on Software Engineering (ICSE), 2024.
  56. C. S. Xia, Y. Wei, and L. Zhang. Automated program repair in the era of large pre-trained language models. In Proceedings of the 45th International Conference on Software Engineering (ICSE 2023). Association for Computing Machinery, 2023.
  57. C. S. Xia and L. Zhang. Less training, more repairing please: revisiting automated program re- pair via zero-shot learning. In Proceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, pages 959-971, 2022.
  58. F. F. Xu, U. Alon, G. Neubig, and V. J. Hellendoorn. A systematic evaluation of large language models of code. In Proceedings of the 6th ACM SIGPLAN International Symposium on Machine Programming, pages 1-10, 2022.
  59. X. Yang, Y. Chen, E. Eide, and J. Regehr. Finding and understanding bugs in c compilers. In Proceedings of the 32nd ACM SIGPLAN Conference on Programming Language Design and Implementation, PLDI '11, page 283-294, New York, NY, USA, 2011. Association for Computing Machinery.
  60. T. Yu, R. Zhang, K. Yang, M. Yasunaga, D. Wang, Z. Li, J. Ma, I. Li, Q. Yao, S. Roman, et al. Spider: A large-scale human-labeled dataset for complex and cross-domain semantic parsing and text-to-sql task. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, 2018.
  61. L. Zhang, D. Marinov, L. Zhang, and S. Khurshid. An empirical study of junit test-suite reduction. In 2011 IEEE 22nd International Symposium on Software Reliability Engineering, pages 170-179. IEEE, 2011.
  62. Q. Zheng, X. Xia, X. Zou, Y. Dong, S. Wang, Y. Xue, Z. Wang, L. Shen, A. Wang, Y. Li, et al. Codegeex: A pre-trained model for code generation with multilingual evaluations on humaneval-x. arXiv preprint arXiv:2303.17568, 2023. · arXiv

Cited by

Sorted by citations

Related

Authors
Liu, J., Xia, C. S., Wang, Y., & Zhang, L.
ORCID
Xia, Chunqiu Steven
Venue
Advances in Neural Information Processing Systems 36
Publisher
Neural Information Processing Systems Foundation, Inc. (NeurIPS)
Published
2023
Pages
21558-21572
ISSN
1049-5258 (ISSN-L)
DOI
10.52202/075280-0943
OpenAlex
W7133198585
Cited by
384
References
0
Citations by year
2026: 143 · 2025: 116 · 2024: 102 · 2023: 23
FWCI
98.69
Type
conference paper
Language
en
Open access
Locations
  • neural information processing systems (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Funders
National Science Foundation (2131943, 2141474)
Topics
  • Physical Sciences › Computer Science › Information Systems › Software Engineering Research
  • Physical Sciences › Computer Science › Software › Software Testing and Debugging Techniques
  • Physical Sciences › Computer Science › Computer Science Applications › Teaching and Learning Programming
Keywords
large language models, code generation, ChatGPT, program correctness, program verification

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review