[Feature]: add support GPT-5.6 on GitHub Copilot

[derrik-fleming]

Check for existing issues

• I have searched the existing issues and checked that my issue is not a duplicate.

The Feature

Summary

Please support direct upstream /responses calls for github_copilot in litellm.responses(...) instead of falling back to the Chat API or a hardcoded endpoint.

Current behavior

With github_copilot, litellm.responses(...) currently routes GPT-5.6 models through the Chat API (/chat/completions) and uses a hardcoded internal endpoint instead of the URL returned during OAuth authentication.

Requested behavior

If the model is a GPT-5.6 variants (or if the endpoint returned during authentication supports /responses), please route litellm.responses(...) directly to the upstream /responses endpoint.

Motivation, pitch

I'm using github_copilot with LiteLLM.

I found that, to call models such as GPT-5.6 on GitHub Copilot, the Responses API is required. However, when I use the github_copilot provider through LiteLLM, I found that litellm.responses(...) does not route to the actual upstream /responses endpoint and instead falls back to Chat API behavior.

Additionally, the endpoint URL is hardcoded and does not use the endpoint returned during OAuth authentication.

I'd like LiteLLM to support direct /responses upstream routing for GitHub Copilot's GPT-5.6 models so that they can be called consistently through the Responses API interface.

What part of LiteLLM is this about?

SDK (litellm Python package)

LiteLLM is hiring a founding backend engineer, are you interested in joining us and shipping to all our users?

No

Twitter / LinkedIn details

No response

[Aayush-engineer]

Dug into the current code — one part of this is already fixed, the other checks out.
Hardcoded endpoint: not accurate anymore. get_complete_url() in responses/transformation.py and messages/transformation.py, and _get_openai_compatible_provider_info() in chat/transformation.py, all already prefer self.authenticator.get_api_base() (the OAuth-returned URL) over the hardcoded default. Might be worth re-testing on latest.
Routing to chat instead of /responses: real bug, but the cause is simpler than a missing feature. github_copilot_supports_responses_api() gates /responses per-model via mode/supported_endpoints in model_prices_and_context_window.json — and there's currently no github_copilot/gpt-5.6 entry at all, so it falls through to "unknown model → chat." Compare gpt-5.3-codex, which is registered with "mode": "responses" and correctly routes native.
So the fix is likely a one-line data addition, not new routing logic. Only open question: is GPT-5.6 on Copilot responses-only (like gpt-5.3-codex), or dual-endpoint (like gpt-5.1/gpt-5.2, which support both but stay on chat)? That determines whether it's mode: "responses" outright or an opt-in via supported_endpoints. Can open a PR once that's confirmed.

[derrik-fleming]

Dug into the current code — one part of this is already fixed, the other checks out. Hardcoded endpoint: not accurate anymore. get_complete_url() in responses/transformation.py and messages/transformation.py, and _get_openai_compatible_provider_info() in chat/transformation.py, all already prefer self.authenticator.get_api_base() (the OAuth-returned URL) over the hardcoded default. Might be worth re-testing on latest. Routing to chat instead of /responses: real bug, but the cause is simpler than a missing feature. github_copilot_supports_responses_api() gates /responses per-model via mode/supported_endpoints in model_prices_and_context_window.json — and there's currently no github_copilot/gpt-5.6 entry at all, so it falls through to "unknown model → chat." Compare gpt-5.3-codex, which is registered with "mode": "responses" and correctly routes native. So the fix is likely a one-line data addition, not new routing logic. Only open question: is GPT-5.6 on Copilot responses-only (like gpt-5.3-codex), or dual-endpoint (like gpt-5.1/gpt-5.2, which support both but stay on chat)? That determines whether it's mode: "responses" outright or an opt-in via supported_endpoints. Can open a PR once that's confirmed.

I tested all three on chat and got:

litellm.BadRequestError: Github_copilotException - model "gpt-5.6-<luna,terra,sol>" is not accessible via the /chat/completions endpoint

Opened #33966 with additions to model_prices_and_context_window.json.