8 views
-/https://github.com/berriai/litellm/issues/27175
GitHub · issue

#27175 [Bug]: Cannot use ChatGPT subscription sucessfully

  • State: open
  • Author: @davewilliamson
  • Labels: bug, proxy, llm translation

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

--- Short Version: --- Confirming this issue with reproducible data.

- LiteLLM `1.83.14`, `chatgpt/*` OAuth device flow - Requests to `https://chatgpt.com/backend-api/codex/chat/completions` return Cloudflare challenge HTML / 403 - Reproduced from two different hosts with same egress IP (so not a single-host routing issue)

Key debug finding: - LiteLLM outbound request includes Authorization + User-Agent - **No Cookie header present** - Challenge HTML is returned instead of JSON

So valid OAuth token alone is not sufficient in this environment for this endpoint path. This appears to be a ChatGPT-provider compatibility gap in LiteLLM under Cloudflare challenge conditions.

--- Longer Version: --- We can reproduce this consistently in a dual-host setup and can confirm it is not just a single-host network anomaly.

## Environment - LiteLLM: `1.83.14` - Provider path: `chatgpt/*` (ChatGPT subscription OAuth device flow)

## Reproduction summary 1. OAuth auth file is present and valid (`access_token`, `refresh_token`, `id_token`, non-expired `expi…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (4 events)
#0 of 0 · 31d18h34m57s ago — entered · #import:https:::github.com:berriai:litellm post #2282
The left issue is harder because it involves diagnosing and adapting to provider-side anti-bot behavior, authentication state, and externally controlled request validation, making reliable implementation and testing uncertain. The right issue is a more bounded gateway data-propagation change, though it still crosses MCP transport and response-handling layers.
#27175 requires deeper provider-authentication changes, external-service compatibility work, and difficult deterministic testing; #32775 is comparatively localized to validation and repository-source handling.
The right issue is harder because it requires investigating and adapting an external provider’s authentication and challenge behavior, with broader integration and regression-testing risk; the left is a localized resource-lifecycle correction with focused validation.
#0 of 0 · 31d18h3m8s ago — current · #import:https:::github.com:berriai:litellm post #2799
Issue 27175 is harder due to broader provider-integration scope, authentication/session handling, and external service behavior; issue 35404 appears comparatively localized to dashboard data propagation and rendering.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search