Our AI-powered code reviews just got a major upgrade. After running evals across 7 models and 10 experiments, we rolled out changes that deliver dramatically better performance across the board.
The numbers speak for themselves: reviews now complete 3.4x faster (4.0s average, down from 13.6s) with 9.6x more consistent latency (standard deviation dropped from 4.8s to 0.5s). That means fewer reviews that feel instant and fewer that feel like they're hanging. Just a reliably fast experience every time. Quality consistency held steady, so you're getting the same thoroughness with a fraction of the wait.
| Metric | Before | After | Improvement |
|---|---|---|---|
| Avg latency | 13.6s | 4.0s | 3.4x faster |
| Latency StdDev | 4.8s | 0.5s | 9.6x more consistent |
| Stress test | 30+s | 5.8s | ~5x faster |
| Cost (output tokens) | baseline | -17% | 17% cheaper |
Quality consistency held steady (StdDev 0.5 in findings), same thoroughness, fraction of the wait. And we reinvest those cost savings back into the platform.
Not using our AI reviews yet? It's super simple to get started. Set up our GitHub or GitLab integration in 2 clicks, or install our CLI for 50+ IDEs with one command. Your coding agents get augmented with long-term knowledge, requirements, guardrails, and on-the-fly relevant context, automatically.