TREX Runs Your Code.

Learn More

Greptile v5

[ Daksh Gupta | 2026-08-05 ]

navigation|BlogGreptile v5
Table of Contents

Today we're releasing Greptile Agent v5, our best code review agent yet. It is faster, better at catching bugs and flags fewer false positives.

We rebuilt Greptile's agent architecture around the idea that agents perform better when their task is narrowly scoped.

Greptile V5 spins up a swarm of agents that each explore one hypothesis for a potential bug. Since these agents run in parallel, our end to end latency is much smaller. Because each agent is focused on a narrowly scoped task, it can explore much deeper when needed, making Greptile better at catching tricky bugs.

We have been A/B testing v5 for the last month across many customers. Here is how v5 does compared to v4 based on live production data from over a million PRs.

  1. Median review time went down from 5:04 to 2:25 seconds.
  2. Percentage of Greptile comments addressed by the author increased from 52% to 66%.
  3. Developers often reply to Greptile's comments. Positive replies ("nice catch" or "fixed") increased by 28.6% from 0.266 per pull request to 0.342 per pull request.
Greptile v4 versus v5 comparison showing more positive replies, more addressed comments, and faster median review time

Greptile v5 is now enabled for all users. We’re excited for you to try it.

Recent feedback

Post from Bence Redmond praising Greptile's accuracy and low false-positive rate
Post from Harsh Vardhan describing Greptile catching bugs in an AI-generated healthcare app
Post from Gui Bibeau describing how Greptile helps get pull requests into good shape
Post from Vikram discussing Greptile T-Rex and sandboxed agent reliability
Post from John Horton praising Greptile's pull request comments
Post from Jinjing Liang showing Greptile in Orca's pull request review process




See Greptile in action