Menu

1 / 2

[Bug]: Codex reasoning restrictions are treated as authentication failures

GitHub Public Timeline Feed·gcp·about 9 hours ago
#WNzs7bNf
Reading 0:00
15s threshold

Bug description

The ChatGPT Codex endpoint can return HTTP 403 with structured code reasoning_effort_not_allowed . Hermes treats it as an authentication failure, although the provider is rejecting the selected reasoning level. This produces misleading credential handling and can silently switch the turn to a fallback model.

Observed response (credentials and conversation content omitted):

{ "error" : { "type" : " invalid_request_error " , "code" : " reasoning_effort_not_allowed " , "message" : " This reasoning level isn't available for the selected model in your workspace. Choose an available reasoning level or another model.…

Continue reading — create a free account

Join HashtagPLUS to read full articles, follow hashtags, vote, and join the conversation.