Anthropic Launches /ultrareview in Claude Code — Multi-Agent PR Analysis with Opus 4.7
Anthropic shipped Ultra Review on April 16 alongside Opus 4.7, bringing cloud-powered multi-agent code analysis to Claude Code. Multiple specialized agents examine PRs in parallel, flag bugs, security issues, and regressions, then verify findings before surfacing them — with 54% of PRs now receiving substantive feedback, up from 16%.
Original sourceAnthropic launched Ultra Review (/ultrareview) in Claude Code on April 16, 2026, shipping it alongside Opus 4.7 as a cloud-powered upgrade to the Code Review feature that debuted in March. Where the original Code Review assigned a single agent to analyze pull requests, Ultra Review deploys a fleet of specialized agents in parallel — each hunting a different class of issue — then runs a verification pass to filter false positives before surfacing findings to the developer.
The system scales with PR complexity. Smaller diffs get lighter review passes; large, sprawling changes across multiple modules get deeper multi-agent treatment. According to Anthropic's internal data, 54% of pull requests now receive substantive comments from Ultra Review, up from just 16% with earlier approaches. Engineers marked fewer than 1% of findings as incorrect — a low false positive rate that distinguishes it from previous automated review tools that developers learned to ignore.
Each review averages $15-25 in API cost, scaling with PR size and codebase complexity, with average review time around 20 minutes. Anthropic is offering Ultra Review free for Claude Pro and Max users through May 5, 2026 as a launch promotion.
The timing is significant. As AI-generated code floods repositories at unprecedented velocity, the need for AI-powered review to catch AI-introduced bugs becomes circular but necessary. Anthropic is positioning Claude Code not just as a code generator but as the full-stack development partner — writing, reviewing, and iterating on code within the same tool.
Ultra Review is available now in Claude Code for Team and Enterprise subscribers, with the Pro/Max free trial running through early May.
Panel Takes
The Builder
Developer Perspective
“Multi-agent PR review that actually verifies findings before surfacing them is the right architecture. The 54% substantive comment rate is impressive — if it holds in production, this is the code review tool most teams have been waiting for.”
The Skeptic
Reality Check
“$15-25 per PR review adds up fast for active teams. And 'AI reviewing AI-generated code' creates a circular quality assurance problem — the same model family writing and reviewing code may share the same systematic blind spots.”
The Futurist
Big Picture
“The code review bottleneck has been a stubborn constraint on developer velocity for decades. If multi-agent verification can reliably replace senior engineers as the review bottleneck, the downstream impact on software team composition and scale is immense.”