Claude Code launches Ultrareview, a multi-agent cloud-based parallel code review to hunt for bugs

According to Dyno Beating monitoring, Anthropic has introduced `/ultrareview` in Claude Code, a cloud-based multi-agent code review feature. Users can input `/ultrareview` in the CLI, and the system will start a set of review agents in a remote sandbox to concurrently inspect the diff between the current branch and the default branch (including uncommitted changes). Users can also input a PR number for a direct review of a GitHub PR. The entire process does not consume local resources, takes about 5 to 10 minutes, and returns the results in notification form within the session.

The key difference between local `/review` and `/ultrareview` lies in the validation mechanism: each finding is independently reproduced and confirmed by an agent, focusing on real bugs rather than code style suggestions. The official documentation positions the two tools as serving different stages: `/review` for quick feedback during the coding process and `/ultrareview` for a deep review of critical changes (such as authentication, data migration) before merging.

On the billing side, `/ultrareview` follows extra usage billing and does not consume in-plan usage. Pro and Max users each have 3 free credits before May 5th (one-time, not refreshed), and thereafter, each review costs about $5 to $20, depending on the scale of changes. Team and Enterprise users do not have free credits. This feature requires a Claude.ai account for authentication and is not available for use in Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, or organizations with zero data retention enabled.