ChatGPT Pro: Spark quota shows 100% remaining, but the model is unavailable and requests return HTTP 400

Summary

I am a ChatGPT Pro subscriber. My account shows 100% remaining in both the five-hour and weekly GPT-5.3-Codex-Spark usage limits, but I cannot use the model.

Spark is missing from the model lists in both the desktop app and Codex CLI. Manually specifying the exact lowercase model name, gpt-5.3-codex-spark, returns HTTP 400 stating that the model is not supported with a ChatGPT account.

The main issue is the mismatch between the displayed Spark quota and actual access: the interface shows an entirely unused allowance, but requests are rejected.

Environment

• Subscription: ChatGPT Pro
• Platform: Windows, PowerShell
• CLI version tested: Codex CLI 0.154.0 (stable release)
• Authentication: ChatGPT account; the same account is used in the desktop app and CLI
• Model: gpt-5.3-codex-spark
• Test date: September 15, 2026

Steps to reproduce

1. Sign in with the affected ChatGPT Pro account.

2. Open the usage page and observe that both Spark usage limits show 100% remaining.

3. Open the model selectors in the desktop app and CLI; Spark is not listed.

4. In PowerShell, start Codex with:

   codex -m gpt-5.3-codex-spark
   

5. Send a short text prompt.

6. Observe the server rejection below.

Actual behavior

The request fails with HTTP 400. The CLI displays:

{"detail":"The 'gpt-5.3-codex-spark' model is not supported when using Codex with a ChatGPT account."}

The CLI also displayed this transport warning before falling back to HTTPS:

Falling back from WebSockets to HTTPS transport. request timed out

After the fallback, the client received the explicit model-not-supported response. The request therefore did not fail solely because of a connection timeout.

Troubleshooting already completed

• Tested the stable release of Codex CLI 0.154.0.
• Used the same ChatGPT Pro account in the desktop app and CLI.
• Refreshed login credentials and tested again.
• Specified the exact lowercase model identifier, gpt-5.3-codex-spark.
• Confirmed that a server response is received over HTTPS.

These checks did not restore Spark access.

Expected behavior and request

Please verify whether my account is eligible to use GPT-5.3-Codex-Spark and confirm which clients or access methods currently support it for my account.

If my account is eligible, Spark should be selectable and requests should succeed while quota remains. If access is restricted or unavailable, the interface should clearly explain that restriction instead of showing an apparently usable 100% remaining allowance without an explanation.

Please investigate the mismatch between the displayed quota, the available model list, and the server-side access check. I cannot determine the underlying cause from the client-side error alone.

Uploaded diagnostic feedback

The CLI confirmed: Feedback uploaded.

Uploaded thread: 01a0a312-f900-73b2-bb8f-699ca34f79a0

Please use this thread ID to locate the submitted diagnostic feedback.

Start by reproducing the request through Codex CLI 0.154.0 in PowerShell with codex -m gpt-5.3-codex-spark, then compare the model selector, quota display, and HTTP 400 response. Use the uploaded diagnostic thread 01a0a312-f900-73b2-bb8f-699ca34f79a0; done means identifying whether the mismatch is client-visible or server-side and documenting the supported access behavior.