Cursor shipped an update on June 10 that cut Bugbot’s average code review time from roughly five minutes to 90 seconds, powered by Composer 2.5. The same update finds 10% more bugs per run (0.62 up from 0.56) and costs 22% less per review, according to Cursor’s changelog. Ninety percent of runs now finish in under three minutes, per Digital Applied’s analysis of the release.
The speed change matters because it shifts where Bugbot fits in a development workflow. A five-minute review is something a team waits for. A 90-second review becomes reflexive enough to run before every push.
Pre-Push Gate, Not Just PR Reviewer
The same release window introduced a /review command that runs Bugbot before a pull request exists. Available in Cursor 3.7+ and on cursor.com/agents, the command syncs with GitHub and GitLab so it does not double-charge when the same diff is later submitted as a PR. Bugbot recognizes the duplicate diff, skips the re-review, and notes it already examined the code, according to Digital Applied.
This moves the check to the cheapest point in the pipeline for catching a bug: before the code leaves the developer’s machine. Shortcut commands /review-bugbot and /review-security let teams target a specific agent. CLI support is listed as coming soon.
Composer 2.5 Under the Hood
The speed and accuracy gains come from Composer 2.5, which Cursor says was trained on 25x more synthetic code review tasks than its predecessor. The model launched on May 18, 2026, and now powers both Cursor’s coding agent and Bugbot.
Benchmark context requires nuance. On CursorBench v3.1, Cursor’s own evaluation, Composer 2.5 scores 63.2%, above Claude Opus 4.7 (61.6%) and GPT-5.5 (59.2%). On the more independent SWE-Bench Multilingual, it lands at 79.8%, roughly 0.7 points behind Claude Opus 4.7 at 80.5%. The Digital Applied analysis notes that CursorBench is a vendor-controlled benchmark, not an independent one. On the independent eval, Composer 2.5 shows near-parity, not superiority.
The cost story is less ambiguous. Cursor positions Composer 2.5 at roughly one-tenth of Claude Opus 4.7’s per-task cost, which is what makes cheaper Bugbot runs feasible at scale.
The ROI Calculation
At Cursor’s stated $1.00 to $1.50 per run and 0.62 bugs found per default review, the cost per caught bug lands in single-digit dollars. Digital Applied cites Stack Overflow and GitHub 2025 data showing repositories using AI-assisted review had roughly 32% faster merge times and 28% fewer post-merge defects compared to human-only review. Against the widely cited four-to-five-figure cost of a production incident, the per-bug economics clear most ROI thresholds.
The update was published by Cursor team members Jason Smale, Yuri Volkov, and Michael Zhao. All performance figures are vendor-stated unless independently noted.