# Support GPT Models for Duo Code Review ## Summary Add OpenAI GPT models as selectable options for Duo Code Review, enabling customers who cannot or prefer not to use Anthropic models to access code review functionality through a GitLab-hosted AI Gateway (AIGW) model. ## Problem Duo Code Review currently only supports Anthropic models (e.g., Claude Sonnet 4.6) as the backend LLM. This blocks customers who are unable to use Anthropic models due to contractual, policy, or procurement constraints. ## Context - GPT 5.2 was previously evaluated through internal dog-fooding and showed promising results for code review quality. - We now have automated evaluation datasets for Duo Code Review, which means we can assess new model candidates significantly faster than before — without relying on weeks-long internal testing cycles. - A partial evaluation of GPT 5.2 exists but a full benchmark run has not been completed. ## Proposal 1. **Complete benchmark evaluation** of GPT 5.2 (and/or other comparable OpenAI models) against the existing Duo Code Review evaluation dataset. 2. **Add qualifying GPT model(s) to the supported model list** for the Duo Code Review flow, available to customers via the GitLab-hosted AIGW. ## Links - Related issue (GPT 5.2 enablement for gitlab-org group): https://gitlab.com/gitlab-org/gitlab/-/work_items/592615