[Bug]: gpt-5.3-codex-spark shown in model selector regardless of account access, leads to API error

[ProRedCat]

When using v0.0.8 (9b9006f), it is possible to select gpt-5.3-codex-spark in the model picker when using API-based billing. This model is not generally available via the API yet, and selecting it results in: {"message":"The requested model 'gpt-5.3-codex-spark' does not exist.","type":"invalid_request_error","param":"model","code":"model_not_found"}.

The model selector UI renders all entries from a static model list without checking whether the model is actually accessible for the current auth type. There is a server-side fallback that downgrades spark to gpt-5.3-codex for ineligible plans, but it only applies to ChatGPT OAuth users. API key users are explicitly treated as spark-enabled where the Codex app-server account/read call returns sparkEnabled: true for API key accounts, so the fallback to GPT 5.3 never happens and the unsupported model is sent directly to OpenAI.

The server already calls model/list on the Codex app-server during session startup but only logs the response. This endpoint returns models filtered by the user's auth type and account access so for API key users it only includes models with supported_in_api: true. This response could be used to populate the model picker with only accessible models, or to validate the selected model before starting a thread.

T3.Code.-.GPT.5.3.Spark.mp4

[shivamhwp]

try in the latest version still happening ?

[ProRedCat]

Fixed by #1371