Today we're releasing Greptile Agent v4, our best code review agent yet. It's better than v3 at catching tricky bugs, and also has a far lower false positive rate.
v3 vs v4 performance
We have been A/B testing v4 for the last month across many customers. After hundreds of thousands of PRs reviewed, we have some interesting real-life data on how v4 compares against v3:
- Addressed comments per PR went from 0.92 to 1.60, a 74% increase. "Addressed" is determined by an LLM.
- Percentage of Greptile comments addressed by the author increased from 30% to 43%.
- Developers often reply to Greptile's comments. Positive replies ("nice catch" or "fixed") increased by 68% from 0.31 per pull request to 0.52 per pull request.
- Upvote reactions on comments increased by 60%, from 0.05 to 0.08 per pull request.
| Metric | v3 | v4 | Change |
|---|---|---|---|
| Addressed comments per PR (determined by LLM-as-judge) | 0.92 | 1.60 | +74% |
| % of comments addressed | 30% | 43% | +44% |
| Positive replies per PR | 0.31 | 0.52 | +68% |
| Upvote reactions per PR | 0.05 | 0.08 | +60% |
Pricing
We're modifying our pricing to a base + usage model, similar to most other AI coding tools. The pricing is $30/developer/month, which includes 50 reviews/month, after which reviews cost $1 each.
Less than 10% of active users will exceed the included usage. Existing users will remain on the existing pricing until the next billing month.
Better coding agents have led to a drastic increase in the number of commits reviewed by Greptile. This new pricing model allows us to focus solely on providing an uncompromising AI code review experience.