Breaking

Anthropic Adds Selectable Effort Levels to Claude Code's Code Review Command

July 17, 2026 at 01:45 EDT

AI Agents
Software Dev & Coding

Anthropic has updated the /code-review command in Claude Code, allowing developers to choose between faster, cheaper reviews and deeper, more exhaustive bug detection. The change, rolled out in mid-July 2026, turns what had been a fixed-depth review process into an adjustable dial that lets teams match the level of scrutiny to the stakes of a given code change.

July 2026 · Claude Code

Code Review Gets a Dial: From Fast & Cheap to Exhaustive

Anthropic's /code-review command now offers five effort levels — each running a distinct review strategy, so teams match scrutiny to the stakes of a change.

16%→54%

Share of pull requests receiving substantive review comments

84%

Large PRs (>1,000 lines) that drew findings — avg 7.5 comments each

<1%

Findings engineers flagged as incorrect

SUBSTANTIVE REVIEW COVERAGE

Portion of PRs getting real findings — before vs after

16%

Before

54%

After

84%

Large PRs
(>1,000 lines)

FIVE EFFORT LEVELS, FIVE STRATEGIES

Each step up changes how the review is run — not just how hard it tries

Low

Single pass over a diff — fastest & cheapest, yet still surfaces more than many tools

Medium

Reads code in context, runs finders from multiple angles, verifies issues before reporting

High

Fresh sub-agents for finding & verification — avoids the bias of the code's author

X-high

Also checks ripple effects on code outside the immediate change

Ultra

A fleet of agents independently reproduces every finding — Anthropic runs it on every internal PR

✓ WHAT DEVELOPERS LIKE

Even Low can outpace competing tools while conserving tokens
High & Ultra improve recall and cut false positives
Eases the human review burden

! THE CAVEATS

Deeper modes are slower and costlier — ~$15–$25/PR, ~20 min per run
False positives not eliminated; human confirmation still needed
Reserve Ultra for merges & high-risk changes

Suggested Workflow

Low

Routine per-push reviews

→

Ultra

Pre-merge gating & high-risk changes (auth, billing)

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.