Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{210545,
        author = {Sonali Shrivastava and Dr. Geetanjali Jindal},
        title = {Benchmarking AI-Generated Code: Evaluating Correctness, Maintainability, and Performance},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {6},
        pages = {1237-1248},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=210545},
        abstract = {AI code generators are becoming more common in software development, but the tests that rate them usually focus only on whether the code works correctly. This article looks at how the correctness; ease of maintenance and speed of AI-generated code are measured. It also asks if these factors are connected and why some studies comparing AI code with human-written code have results. A detailed summary of 16 sources published from 2023 to 2026 was made. These sources included five surveys, four studies that tested models, six small studies and theses and one benchmark paper. Some of the numbers from the studies were recalculated. The summary found that (a) passing rates on tasks are high but don’t say much about other qualities, (b) there is no standard way to measure maintainability and the tools used to check it don’t agree much (c) AI code that works correctly can be much slower than code written by experts (d) whether AI code is better or worse, than human code depends on the human example used, the programming language, the tools and how results are adjusted and (e) fixing AI code in steps can add new problems. Based on these findings the article suggests a way to test AI code. This includes a metric that checks quality, a set of things to report and a way to test the method. The results are for the sources used and the new way has not been tested in real situations yet.},
        keywords = {AI-generated code, large language models, code generation benchmarks, software quality, maintainability, static analysis},
        month = {November},
        }