Spark allowance shows 100% remaining but prolite account receives model-not-supported HTTP 400

Summary

My account reports a GPT-5.3-Codex-Spark allowance with 100% remaining in both the five-hour and weekly windows, but Spark is missing from the desktop model picker and an explicit CLI request returns HTTP 400.

Environment

• Windows 11, OS build 26200, PowerShell
• Codex CLI: 0.154.0-alpha.6.2 (desktop-bundled binary)
• Authentication: ChatGPT account
• Account plan identifier returned by usage reporting: prolite
• Model requested: gpt-5.3-codex-spark
• Reproduced: September 16, 2026

Steps to reproduce

1. Check account usage. A separate limit named GPT-5.3-Codex-Spark reports usedPercent: 0 for both the 300-minute and 10080-minute windows.
2. Open the desktop model picker. It offers 5.6 Sol, 6 Astra, 5.6 Terra, 5.6 Luna, and 5.5, but no Spark.
3. Run:

codex exec --model gpt-5.3-codex-spark --sandbox read-only --skip-git-repo-check --ephemeral --color never 'Reply with exactly: Spark connection works. Do not use tools or read or edit files.'

Actual result

The CLI exits with code 1. It first warns:

Model metadata for `gpt-5.3-codex-spark` not found. Defaulting to fallback metadata; this can degrade performance and cause issues.

The server then returns:

{"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.3-codex-spark' model is not supported when using Codex with a ChatGPT account."}}

This is an explicit model-access rejection, despite the unused allowance. I have not established the underlying cause or tested a separate stable CLI installation.

Expected result

If the account is eligible, Spark should be selectable and direct requests should work. If it is ineligible, the usage interface should explain why it shows a Spark allowance that cannot be used. Please clarify Spark eligibility for the prolite plan and investigate the mismatch between usage reporting and model access.

Related reports

• https://github.com/openai/codex/issues/45594
• https://github.com/openai/codex/issues/45972

This report adds a reproduction using the desktop-bundled CLI version above and the plan identifier returned by usage reporting. Account IDs, credentials, local personal paths, and unrelated logs are omitted.

Update after reviewing related reports (September 16, 2026)

I reviewed #45594, #45972, #45857, #45613, and #45728 after the duplicate notice. They describe the same basic problem across Windows and macOS: Spark quota is visible, but the model is missing or direct requests are rejected. Reports using stable CLI 0.154.0 also reproduce it, and users in #45594 and #45613 say signing in again did not help. All five issues were still open when checked, with no maintainer explanation or confirmed fix in their comments. The author of #45857 reports a support escalation, but no diagnosis yet.

I understand why these reports are flagged as possible duplicates. What feels unfair is being shown an available allowance without being able to use it or getting a clear explanation. The duplicate notice points to other people with the same problem, but it does not answer why access is being rejected.

Can someone from OpenAI clarify whether Spark is currently supposed to work for these accounts? If it is, is there a known incident or a supported fix? If it is no longer available, why does the usage page still show an unused Spark allowance? Please identify the main issue to follow and provide a clear status update. I do not know the cause and am asking for an explanation rather than assuming one.

Start with the reported Codex CLI invocation and compare its behavior with the related issues #45594, #45972, #45857, #45613, and #45728. The issue is done when Spark eligibility for the prolite plan and the mismatch between usage reporting and model access are clearly explained or a confirmed fix is identified.