gpt-5.6-sol' model is not supported and no usage limit

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #34918
• #35923

Powered by Codex Action

I can reproduce this on a renewed ChatGPT Pro ($200 / 20×) subscription.

Environment

• Codex CLI: v0.146.0
• Authentication: ChatGPT account login
• Subscription: ChatGPT Pro ($200 / 20×), recently renewed after the previous subscription expired
• Model selected: gpt-5.6-sol

What happened

Immediately after renewing the subscription, selecting Sol in Codex CLI returned:

The 'gpt-5.6-sol' model is not supported when using Codex with a ChatGPT account.

Signing out and signing back in did not fix it. Codex then switched the model from Sol to Terra and rejected the request with a usage-limit error:

You've hit your usage limit. Add credits to continue, or try again at Aug 8, 2026 11:41.

However, the account still has usage available, and the same ChatGPT account can use gpt-5.6-sol successfully in ChatGPT's web Work mode at the same time.

Expected behavior

After a Pro renewal, Codex CLI should recognize the refreshed subscription/model entitlement and remaining usage consistently with ChatGPT Work mode. Logging out and back in should refresh the entitlement if it has become stale.

This looks like a cross-client entitlement/usage synchronization issue triggered by subscription expiration followed by renewal, rather than an actual usage-limit exhaustion.

Same symptom on ChatGPT Pro, reproduced 2026-08-07.

Environment

• auth_mode: chatgpt (no API key configured)
• chatgpt_plan_type = pro, subscription active and not expired
• Codex CLI: reproduced on both 0.144.1 (Homebrew) and 0.147.0-alpha.6.5 (bundled with the desktop app)

Symptom

codex exec -m gpt-5.6-sol → HTTP 400

The 'gpt-5.6-sol' model is not supported when using Codex with a ChatGPT account.

gpt-5.6-terra and gpt-5.6-luna succeed with the exact same command, auth and config.
The model name is already the variant form, so this is not the plain gpt-5.6 case.

Ruled out (each reproduced the identical 400):

• Ran with no local config at all (moved ~/.codex/config.toml and ~/.codex/models.json aside)
• Re-authenticated from scratch (sign out / sign in, fresh id_token)
• Both CLI versions above
• No project-level .codex/config.toml present

Inconsistency worth noting: ~/.codex/models_cache.json, re-fetched minutes earlier under this
same ChatGPT auth, lists gpt-5.6-sol with visibility: "list" and supported_in_api: true.
The model-listing API advertises Sol for this account while the execution API rejects it.

As in the original report, this account was upgraded to Pro shortly before the first failure,
which points at entitlement propagation rather than anything client-side.

Same issue.

Same issue，how to resolve it？