Urgent.News

What's breaking now, across thousands of outlets.

AI

Open Code Review: How Alibaba Open-Sourced an AI Code Reviewer That Cuts Token Costs by 90%

Code review is one of the highest-leverage practices in modern software engineering, yet it remains one of the biggest bottlenecks. In high-velocity teams, pull requests sit idle waiting for senior developers to triage them, while junior reviews often get caught up in formatting nitpicks rather than deep architectural bugs. When general-purpose AI coding agents (such as Claude Code or Cursor)…

Alibaba has open-sourced a powerful AI code reviewer called Open Code Review (OCR) that can significantly reduce token costs by 90% compared to traditional general-purpose AI coding agents. This tool is designed to address three major pain points in high-velocity teams: incomplete code coverage, position drift, and runaway token costs.

OCR takes a hybrid approach, combining deterministic engineering with AI agents to handle different aspects of code review. It uses deterministic code to accurately select and bundle files for review, ensuring none are missed. Then, specialized AI agents perform semantic reasoning on these bundles to provide accurate, line-level comments and identify defects.

The architecture of OCR avoids the pitfalls of purely text-generating AI by separating concerns. Deterministic pipelines ensure precise file selection, smart bundling of related files, and accurate rule matching based on file characteristics. Positioning modules validate the coordinates of AI-generated comments against the actual code, preventing position drift. A curated toolset for the AI agents retrieves only the necessary context to verify bugs, minimizing unnecessary token usage.

To benchmark OCR against a generic AI agent, Alibaba created the AACR-Bench, a real-world code review test involving 50 open-source repositories, 200 pull requests, and 1,505 verified defects from senior engineers. The results showed OCR significantly outperformed the generic agent in precision, F1 score, and token usage. It achieved an 89% reduction in API costs and provided much faster review speeds compared to the generic agent.

To use OCR, developers can install the cross-platform CLI tool globally with npm. The CLI supports various model providers, including OpenAI, Anthropic, and self-hosted local models. Once configured, OCR can be run on git diffs to generate accurate code review comments directly in the terminal or CI pipelines. This open-source solution brings the benefits of AI-powered code review to teams of all sizes, minimizing the cost and increasing the effectiveness of the software development process.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

OpenAI takes its AI fight with Anthropic to Big Law

OpenAI launches Astra for Law, targeting AmLaw 200 firms with advanced legal AI tools. It's a bid to outpace Anthropic in the rivalry over legal work.

  • OpenAI launches Astra for Law to compete with Anthropic.
  • Integrates with Harvey and Legora legal tech firms.
  • Features plugin tools and enhanced legal case search.

UN turns to Google to make its global data ready for AI agents

The shift comes after a UNICEF test found leading AI models struggled to accurately retrieve global development statistics.

  • UN partners with Google to create UN System Data Commons platform.
  • Supports Model Context Protocol for AI to connect to external data.
  • UNICEF benchmark shows average accuracy of 21.2% for global development indicators.

More from Thursday 17 September →