Residence
Newest reporting
Information
Alibaba’s Open Code Assessment, a hybrid AI-powered code assessment instrument, reaches 51,009 stars on GitHub, showcasing scalability and precision.
On November 14, 2026, Alibaba’s open-source venture Open Code Assessment surged on GitHub Trending, gaining 897 stars in a single day to achieve a complete of 51,009 stars and three,804 forks. Written primarily in Go, this hybrid AI-powered code assessment instrument has been described as safe, quick, environment friendly, and battle-tested at Alibaba’s scale.
Technical Options and Capabilities
Open Code Assessment combines deterministic pipelines with an LLM (giant language mannequin) agent to ship exact line-level feedback throughout code opinions. It features a built-in multi-language ruleset addressing points akin to null pointer exceptions (NPE), thread-safety, cross-site scripting (XSS), and SQL injection vulnerabilities. The instrument is suitable with OpenAI and Anthropic fashions, offering flexibility for builders to configure their most popular mannequin endpoints.
The instrument helps each diff-based opinions for pull requests and full-file scans for auditing unfamiliar codebases. By leveraging its hybrid structure, Open Code Assessment balances deterministic engineering with the contextual understanding of an LLM agent, making certain each precision and depth in its evaluation.
Benchmark Efficiency
In response to benchmarks performed on a dataset of fifty common open-source repositories, 200 actual pull requests, and 10 programming languages, Open Code Assessment demonstrated considerably larger precision and F1 scores in comparison with general-purpose brokers like Claude Code. Notably, it achieved these outcomes whereas consuming solely about one-ninth of the tokens and finishing opinions quicker. Nevertheless, its recall was deliberately decrease, prioritizing precision to cut back false positives.
The benchmark dataset, often known as AACR-Bench, was cross-validated by over 80 senior engineers and included 1,505 annotated ground-truth points. Metrics akin to precision, recall, F1 rating, common assessment time, and token consumption have been evaluated to focus on the instrument’s effectivity and cost-effectiveness.
Adoption and Sensible Use
Initially developed as an inside instrument at Alibaba, Open Code Assessment has recognized hundreds of thousands of code defects over two years of use by tens of 1000’s of builders. Its open-source launch goals to profit the broader developer neighborhood. Set up is easy by way of npm, and the instrument integrates seamlessly with Git for diff technology and repository operations. Builders can use instructions like ocr assessment for pull request opinions or ocr scan for full-file audits.
As of November 14, 2026, the repository is licensed underneath Apache License 2.0 and continues to achieve traction amongst builders worldwide. For extra particulars, customers can go to the venture’s official web site at open-codereview.ai.
