Anthropic has officially introduced effort levels to the /code-review command in all Claude Code sessions. This update allows developers to actively select the level of source code analysis that fits their token budget and the security requirements of each pull request (PR). This is a practical move from Anthropic to balance the cost of running large language models (LLMs) with the quality of code review.
Background & Causes
Previously, automated AI code reviews often applied a fixed analysis pipeline, leading to wasted token resources on minor edits, or conversely, insufficient depth for major architectural changes. According to the @ClaudeDevs development team at Anthropic, dividing the review into different effort levels helps developers decide their own tradeoff. The new feature is now available across all Claude Code sessions when users update the tool to the latest version and run /code-review on their next PR.
Technical Analysis & Technology
The new code review feature is designed with three main effort levels: Low, High, and Ultra. At the "Low" level, the system optimizes costs by minimizing token consumption, yet Anthropic claims it still outperforms other code review tools in finding bugs. The "High" level aims for a significantly higher recall rate when users need to go deeper into system details. Most notably, the "Ultra" level (/code-review ultra) operates by spawning a fleet of reviewer agents that work independently to reproduce and verify every finding, minimizing false positives while maintaining the same severe-issue coverage.
Expert Opinions & Outlook
Representatives from the Anthropic development team stated that they have applied this supreme /code-review ultra level to every internal PR at the company to ensure system stability and safety before practical deployment. Industry observers note that integrating an independent network of agents to self-reproduce bugs is a smart approach, reducing the manual review burden on QA engineers and lead developers.
Impact & Future
This change provides the global developer community with a more flexible tool to optimize Anthropic API usage costs. Instead of paying a large fixed fee for each code scan, startup projects or individual developers can choose the "Low" level for daily tasks and only trigger the "Ultra" mode before merging source code into the main branch. This trend of smart task delegation is expected to soon become a standard for AI-integrated programming tools in the near future.