BREAKING
Claude Code Adds Review Effort Levels
Five Levels From Low to Ultra
1
Low: single pass
↓
2
Medium: in-context
↓
3
High: fresh agents
↓
4
X-high: ripple check
↓
5
Ultra: agent fleet
0
%
PRs before
0
%
PRs after
0
%
large PRs flagged
0
$
max per PR
0
min
typical run
0
x
comments large PR
When to Use Which Mode
Low
routine
●
Fast and cheap
●
Everyday pushes
Ultra
high-risk
●
Independent verification
●
Auth or billing changes
Research Preview on Team and Enterprise
AI NEWS BLITZ
Anthropic now lets developers dial the depth of Claude Code's code reviews.