Cursor Bugbot Gets 3× Faster + Pre-Push Review — Catch Bugs Before They Hit CI

Cursor's June 10 Bugbot update delivers three concrete metric improvements alongside two new workflow features that shift where in the development cycle code review happens. The update is available in Cursor 3.7+ and at cursor.com/agents.

The three headline metrics. Bugbot is now 3× faster overall — 90% of runs now complete in under 3 minutes, down from roughly 5 minutes previously. Each run is 22% cheaper per review, a direct reduction to automated-review cost at any scale. And Bugbot now finds 10% more bugs per review pass, meaning the quality bar rose alongside the speed and cost improvements. Cursor attributes all three gains to harness engineering improvements combined with the Composer 2.5 model that now powers Bugbot — the same model that positions Cursor against GitHub Copilot as a vertically integrated coding environment.

Pre-push review: the workflow shift. The most substantive new feature is pre-push review. Developers can now run the /review command (or the aliases /review-bugbot and /review-security) locally inside Cursor before pushing — Bugbot analyses the diff and surfaces bugs, security issues, and code-quality problems before the branch ever reaches CI. This is a meaningful shift: historically, Bugbot was a post-push, PR-triggered tool. Pre-push review moves the detection point earlier, where fixes are cheaper and faster — no CI queue, no PR comment loop, no waiting for a teammate to triage. Bugbot syncs with GitHub and GitLab to skip re-analysing already-reviewed diffs, so running /review locally before push doesn't produce duplicate comments when the PR opens.

Incremental review. The second new feature is incremental review — Bugbot can now review only the delta since its last review pass rather than re-analysing the full PR on every push. For long-lived PRs that accumulate many commits, this cuts review time and cost substantially. The combination of incremental review and the baseline 3× speed gain means that even large, iterative PRs now get near-real-time feedback on each new commit.

Composer 2.5 as the engine. Cursor explicitly links the quality and speed improvements to Composer 2.5, its in-house frontier coding model trained on Cursor's own post-training pipeline. The implication is that Bugbot's review quality will continue to improve as the underlying model is updated — unlike tools that rely on a fixed third-party model API. For teams comparing options, Cursor vs Claude Code covers where the GUI-first and terminal-native approaches differ on agentic review workflows.

CI economics and team impact. Sub-3-minute Bugbot runs change the economics of automated code review. At 5 minutes, Bugbot was async — you pushed, Bugbot ran, you checked results later. At under 3 minutes, and with pre-push running locally, Bugbot is close enough to real-time that it fits inside a normal commit-review loop. The 22% cost reduction compounds at scale: a team running Bugbot on 100 PRs a month saves meaningfully across the year, and those savings justify expanding review coverage to branches that previously weren't worth the cost. Engineering teams that have been deferring Bugbot adoption on cost or latency grounds now have fewer reasons to wait.