GPT-5.3-Codex-Spark works via explicit CLI invocation but is absent from model catalogue and selectors

What version of the Codex App are you using (From “About Codex” dialog)?

26.908.9136.0

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Microsoft Windows NT 10.0.19045.0 x64

What issue are you seeing?

I am a ChatGPT Pro user with a separate GPT-5.3-Codex-Spark usage allowance visible on my account.

GPT-5.3-Codex-Spark does not appear in any of the model selectors I have tested:

• Codex Windows Desktop
• Codex VS Code extension
• Codex CLI /model

However, Spark works successfully when explicitly invoked from the CLI:

codex -m gpt-5.3-codex-spark

The resulting session starts successfully and reports:

OpenAI Codex (v0.154.0)
model: gpt-5.3-codex-spark high

This suggests that my account/backend permits use of Spark, but Spark is not being advertised in the model catalogue supplied to my clients.

I updated the Codex CLI to the current npm release and confirmed:

codex --version
codex-cli 0.154.0

I also ran codex doctor. The installation is reported as consistent, ChatGPT authentication is configured, OpenAI endpoints are reachable, and the Windows Codex sandbox has been successfully configured.

Running /model in the CLI currently shows:

gpt-6-astra
gpt-5.6-sol
gpt-5.6-terra
gpt-5.6-luna
gpt-5.5

Spark is absent.

I then ran:

codex debug models

The raw catalogue also contains no gpt-5.3-codex-spark entry.

I additionally inspected .codex\models_cache.json. At the time of testing it showed:

"fetched_at": "2026-09-15T10:39:15.082817934Z"
"client_version": "0.154.0"

The model slugs in that freshly fetched catalogue were:

gpt-6-astra
gpt-reserve
gpt-5.6-sol
gpt-5.6-terra
gpt-5.6-luna
gpt-5.5
codex-auto-review

Again, there is no gpt-5.3-codex-spark entry.

So the behaviour appears to be:

1. My Pro account displays a separate Spark usage allowance.
2. Direct CLI invocation of Spark succeeds.
3. A freshly fetched model catalogue does not contain Spark.
4. Consequently Spark is absent from Desktop, VS Code, and CLI /model.

Expected behaviour: because the account has Spark access and direct invocation succeeds, I would expect Spark to be returned in the model catalogue and therefore appear in the normal model selectors.

Could this be an account entitlement / rollout / model-catalogue synchronization issue?

What steps can reproduce the bug?

1. Sign into Codex using a ChatGPT Pro account that has a separate GPT-5.3-Codex-Spark usage allowance.

2. Open the Codex Windows Desktop model selector. GPT-5.3-Codex-Spark is absent.

3. Check the Codex VS Code extension model selector. GPT-5.3-Codex-Spark is also absent.

4. Run Codex CLI 0.154.0 and enter:

   /model

GPT-5.3-Codex-Spark is absent from the model list.

5. Exit Codex and explicitly launch Spark:

   codex -m gpt-5.3-codex-spark

The session starts successfully and reports:

model: gpt-5.3-codex-spark high

6. Inspect the raw model catalogue:

   codex debug models

GPT-5.3-Codex-Spark is absent.

7. Inspect .codex\models_cache.json.

The cache is freshly fetched using client version 0.154.0, but contains no gpt-5.3-codex-spark model entry.

What is the expected behavior?

GPT-5.3-Codex-Spark should be included in the model catalogue and appear in the Codex Desktop, VS Code and CLI /model selectors when the ChatGPT Pro account has Spark access.

The account appears to have valid Spark access because it has a separate Spark usage allowance and explicit CLI invocation of gpt-5.3-codex-spark succeeds.

Additional information

This does not appear to be a stale local cache issue.

At the time of testing, .codex\models_cache.json showed:

fetched_at: 2026-09-15T10:39:15.082817934Z
client_version: 0.154.0

The freshly supplied catalogue contains Astra, Sol, Terra, Luna, GPT-5.5 and hidden/internal models, but no Spark entry.

codex doctor also reports the installation as consistent, ChatGPT authentication configured, OpenAI endpoints reachable, and the Windows sandbox successfully configured.

The key diagnostic distinction is that direct model invocation succeeds while model discovery/catalogue listing does not advertise Spark.

Start with the CLI model-discovery paths exposed by /model and codex debug models, then inspect how .codex\models_cache.json is fetched and consumed. Reproduce with Codex CLI 0.154.0 and an entitled Pro account; done means the catalogue includes gpt-5.3-codex-spark and it appears in the normal model selectors while explicit invocation continues to work.