GitHub Copilot API (api.githubcopilot.com) returns a limited model list for OAuth Apps, even though premium models are enabled on my account

[stuti-khare-maersk]

🏷️ Discussion Type

Question

💬 Feature/Topic Area

API

Body

Hi team,

I have a GitHub Copilot premium subscription and can see multiple premium models (GPT-4.1, GPT-5, GPT-5.1, etc.) available in my IDE (IntelliJ) Copilot Chat model picker, which I use regularly.

I've built a custom internal application that authenticates via a GitHub OAuth App (scope=copilot) to call the Copilot chat completions API on behalf of the signed-in user. When I query the model catalog with that token:

GET https://api.githubcopilot.com/models
Authorization: Bearer <user's copilot token>
Copilot-Integration-Id:
...the response only includes these models, regardless of what's enabled in my personal Copilot settings:

gpt-4o-mini-2024-07-18, gpt-4o-2024-11-20, gpt-4o-2024-08-06,
gpt-3.5-turbo-0613, gpt-3.5-turbo, gpt-4o-mini, gpt-4o
To confirm this isn't a naming/config mismatch on my end, I explicitly requested a premium model (e.g. GPT-5.5) directly in a /chat/completions call — it was rejected with the same limited list above, unchanged.

My question: Is model availability for third-party OAuth Apps / Copilot integrations governed separately from the models enabled for a user's first-party Copilot Chat surfaces (IDE, github.com)? If so:

Is this gated by something on my OAuth App's configuration (e.g. needing formal registration/review as a Copilot Extension), or is it an account/org policy setting?
Where can I request or enable access to the newer models (GPT-4.1, GPT-5.x) for this integration?
This is blocking a production tool for my team, so any pointers — even just confirming which of the above is the cause — would help a lot. Happy to share more request/response details if useful.

Thanks!

[rohanjoshi7]

Hi there.

This is a known architectural limitation with how GitHub handles Copilot API access for custom integrations.

You mentioned seeing models like GPT-4.1 and GPT-5 in your IDE but only getting a limited list like gpt-4o and gpt-3.5-turbo when querying the API. The short answer is yes, model availability for third party OAuth apps is governed entirely separately from the models enabled for first party Copilot Chat surfaces.

First party surfaces like IntelliJ use specific, internal client IDs that GitHub recognizes. These official clients receive access to premium or preview models based on your user and organization entitlements. When you authenticate via a custom OAuth App using the copilot scope, you are routed to a standard API tier. GitHub restricts this token tier to a specific whitelist of stable models. They do this to manage backend costs and prevent rate limit abuse from automated custom scripts.

To answer the specific questions in your post:

• This is not an organization policy setting you can simply toggle. It is a strict restriction on the API side for standard OAuth tokens.

• Formal registration as a Copilot Extension is indeed the intended route for custom integrations. However, even official Copilot Extensions are currently limited to the specific models GitHub supports for the extension ecosystem. That ecosystem usually lags behind the experimental models available in first party IDEs.

• Because you are querying api.githubcopilot.com directly, which is technically an undocumented endpoint for public developer use, there is no official support form or channel to request a model whitelist update for your specific OAuth token.

If this is blocking a production tool for your team, relying on the undocumented api.githubcopilot.com endpoint is risky because the token scopes and routing can change without notice. The best path forward is to build a proper GitHub App configured as a Copilot Extension. You can find the official guides on the GitHub Developer documentation portal. While you might still be restricted to stable models like GPT-4o for now, it is the only officially supported path and will be the first place GitHub officially rolls out newer models for third party use.

[stuti-khare-maersk]

Hi Roshan,

Thanks for the clarity on the OAuth-app model whitelist — that answers the "why" for us.

For context on our use case: we have an internal web app (Spring Boot backend) where each engineer connects their own GitHub account via OAuth (copilot scope), and the app then calls the Copilot chat completions API on their behalf to generate QA test cases from internal design documents. So this is a "bring your own Copilot seat" pattern — we want each user's own entitlements (including whatever premium models their individual/org plan grants them in the IDE) to carry over, not a shared/fixed model for everyone.

Given that:

Custom OAuth apps are capped to the stable model whitelist regardless of the user's actual IDE entitlements, and
Copilot Extensions are currently capped to the same stable models too,
what is the officially supported way to build this kind of per-user tool today? Specifically:

Is there any registration path (Copilot Extension or otherwise) where the model list returned reflects the authenticated user's actual entitlements rather than a fixed extension-wide whitelist?
Is GitHub Models a separate, more appropriate API for this kind of third-party use case, or is that also capped the same way?
If neither gives per-user premium access today, is your recommendation to just stay on gpt-4o via the OAuth flow we have, and treat Copilot Extension migration as forward-looking only (in case per-user entitlement passthrough is added later)?
We'd rather build against whatever GitHub considers the "correct" long-term integration point now, even if it's not better than gpt-4o today, than build further on the undocumented endpoint. Appreciate any pointers to docs or examples of teams doing per-user model passthrough.