Claude 3.7 Sonnet Code Review Summary rollout plan

Overview

We should update the Code Review Summary feature to use the latest available models from Anthropic (Sonnet 3.7 as of this writing).

Related to #521034 (closed).

Resource Links
Model https://www.anthropic.com/news/claude-3-7-sonnet
Epic or Issue #521034 (closed)
Feature Flag Rollout Issue #522987 (closed)
Status updates

Rollout success criteria

Add a list of success criteria here

Dashboard References

This can be the acceptance rate or latency dashboards filtered to the new model. Add as many dashboards as is relevant.

https://log.gprd.gitlab.net/app/dashboards#/view/f959393c-82c1-4b69-a4d3-2446aab9476c?_g=(refreshInterval%3A(pause%3A!t%2Cvalue%3A60000)%2Ctime%3A(from%3Anow-7d%2Cto%3Anow))

Legal notes

Add legal notes here

Known issue list

List of issues identified throughout the evaluation, implementation, and rollout of the model.

Rollout

Timeline

Optional: add a short description here of the expected timeline.

Date Audience Status
2025-03-05 Code Review team members and other stakeholeders
2025-03-06 All GitLab team members
2025-03-07 100% of all users

Feedback from GitLab team members

Add link to the internal feedback issue.

Persevere / Continue Criteria

  1. Latency remains within observed p50/90/95 ranges below
  2. Error rate remains within observed range below, or improves
  3. Nothing was raised as a blocker

Observed latency from May 17 to Aug 21

  • p50: X
  • p90: X
  • p95: X

Observed error rate from July 4 to Aug 21

  • X %

Pivot / Pause / Rollback Criteria

  1. Requests are not using the new model as expected
  2. Opposite of what's defined in the previous section

Mitigation and Rollback Plan

We will use a Feature Flag to control the rollout. If there are any concerns (see above), we will disable the feature flag, especially for external users, to investigate any potential issues.

Release Announcement

Add details here about where to make announcements when the model is ready for rollout to external users.

Edited by Kinshuk Singh