Environment
- Surface: Codex desktop app and Codex CLI on Windows, using ChatGPT sign-in—not an API key.
- Endpoint or feature: Model availability and Codex-Spark usage allowances.
- Model:
gpt-5.3-codex-spark - SDK and version: Desktop-bundled Codex CLI
0.154.0-alpha.6.2. - Subscription: Web profile displays PRO; CLI
/statusdisplays Pro Lite.
Bug
Both Codex-Spark allowance meters—five-hour and weekly—show 100% remaining, but Spark is missing from the desktop model picker and interactive CLI /model list.
Explicitly requesting Spark through the CLI is rejected by the server.
I expected either access to the model whose allowances are displayed, or a clear explanation that my subscription/client is not eligible. I’m unsure whether this is an entitlement problem or misleading usage reporting.
Signing out and back in did not change the desktop model list.
This is what available:
Reproduction
- Sign in to Codex using a ChatGPT account.
- Run
/statusin an interactive CLI session. Observe unused Spark allowances. - Run
/model. Available models are:gpt-6-astragpt-5.6-solgpt-5.6-terragpt-5.6-lunagpt-5.5
- Explicitly request Spark:
codex exec --ephemeral --ignore-user-config --skip-git-repo-check --sandbox read-only --model gpt-5.3-codex-spark -c model_reasoning_effort=medium --json "Do not use tools or inspect files. Reply exactly: SPARK_ACCESS_OK"
- The request fails before generating a response.
Diagnostics
- Exact error and HTTP status: Server error event reports status 400, type
invalid_request_error:
The 'gpt-5.3-codex-spark' model is not supported when using Codex with a ChatGPT account.
- Additional CLI warning:
Model metadata for `gpt-5.3-codex-spark` not found.
Defaulting to fallback metadata; this can degrade performance and cause issues.
- Request ID: No request ID was exposed in the captured output.
- When: 15 September 2026, approximately 17:05 BST / 16:05 UTC.
- Frequency: One explicit CLI generation test captured; Spark remained absent across desktop and interactive CLI checks. Desktop availability remained unchanged after signing out and back in.
- Other checks:
codex login statusconfirms ChatGPT authentication.codex doctorreports consistent installation, configured authentication, successful WebSocket connectivity and reachable provider endpoints. It also reports aTERM=dumbterminal-display limitation in the automated shell and a general Defender warning. - Support status: The issue has been escalated to a support specialist; investigation is pending.


