OpenAI Codex gpt-5.4 'Not Supported' Error
What You'll Learn
- How to distinguish lifecycle and entitlement issues from temporary incidents
- How ChatGPT sign-in differs from API key sign-in for Codex access
- A safe, reversible checklist to clear common configuration pitfalls
- Which supported models to select when gpt-5.4 is unavailable
OpenAI Codex gpt-5.4 Not Supported Error: What It Means
If Codex returns an availability or not supported message when you request gpt-5.4, the platform is telling you the model cannot be used from your current context. The most common reasons include lifecycle changes to specific models, workspace or account entitlements, invalid saved settings, or a transient service condition that blocks the request.
Begin by gathering specifics. Note your sign-in route, whether you are using the ChatGPT sign-in or an API key. Identify the surface where you see the error, such as the desktop app, Codex CLI, IDE extension, or Codex cloud chat. Capture the exact model string, for example gpt-5.4 or gpt-5.4-mini, and the date and time of the failure. If applicable, add your CLI or app version and any workspace policy that might restrict models.
Older reports of outages or routing quirks may describe a different product state. The current path to resolution relies on up to date documentation and the product changelog. You can consult the official Codex Models page at https://developers.openai.com/codex/models and the Codex Changelog at https://developers.openai.com/codex/changelog for current guidance.
Why Codex Shows a Model Availability Error
Codex validates your requested model against your authentication route and surface. Even if a model exists in the OpenAI API, it can be unavailable in Codex when you are signed in with ChatGPT, or restricted by workspace policy on a given surface. In other cases, a saved model string in configuration or an unsaved switch in the UI can keep pointing the tool to a model that is not eligible.
Use this quick classification to map the error to a likely cause and a next step.
| Symptom | Likely Cause | Next Step |
|---|---|---|
| Model gpt-5.4 not available only with ChatGPT sign-in | Lifecycle or entitlement difference for ChatGPT sign-in | Check the Codex Models page and consider a supported replacement |
| Model absent in a managed workspace but available in personal | Workspace policy blocks that model | Ask an admin or use an approved model in the workspace policy |
| CLI says current model is gpt-5.4 despite switching in UI | Saved config still points to gpt-5.4 | Inspect and change the model entry in config.toml |
| Error disappears after re-auth | Stale cached credentials | Logout and sign in again safely, without deleting credential files |
When in doubt, verify model availability and product scope in the official docs before you assume a temporary incident.
ChatGPT Sign-In vs API Key Sign-In
Codex supports two OpenAI sign-in methods. The ChatGPT sign-in grants access based on your ChatGPT plan and workspace entitlements. The API key sign-in grants access based on API model availability and is billed at OpenAI Platform API rates. Some ChatGPT workspace and Codex cloud features can be limited or unavailable when you use an API key.
In local tools, such as the ChatGPT desktop app, Codex CLI, and the IDE extension, you can choose either sign-in route. Codex cloud currently requires ChatGPT sign-in. Details are documented at the official Authentication page: https://developers.openai.com/codex/auth.
| Route | Where It Works | Billing and Access | Notes |
|---|---|---|---|
| ChatGPT sign-in | Desktop app, CLI, IDE extension, Codex cloud | Access based on ChatGPT plan and workspace entitlements | Some models can be retired on this route while API remains available |
| API key sign-in | Desktop app, CLI, IDE extension | Usage billed at standard API rates | Codex cloud requires ChatGPT sign-in, some features may be limited with API key |
Never paste, publish, or share API keys. Treat your cached credentials as sensitive. Codex can store cached login details in ~/.codex/auth.json or in your OS credential store. Logging out in one local tool signs you out for the next run in other local tools that share the cache.
The August 31 2026 GPT-5.4 Codex Retirement
The Codex Models page states that GPT-5.4 and GPT-5.4 mini retire from Codex when you sign in with ChatGPT on August 31, 2026. It also states that this retirement does not affect the OpenAI API or Codex sessions authenticated with an API key. If you continue to use Codex with ChatGPT sign-in after that date, switch to a supported model.
The page recommends the following replacements in saved settings and managed configurations:
- Replace gpt-5.4 with gpt-5.6-terra
- Replace gpt-5.4-mini with gpt-5.6-luna
Where to update these values:
- Workspace defaults and saved model settings
- Managed configurations for teams
- Custom agents and scheduled tasks
- Local configuration in the desktop app, Codex CLI, and IDE extension using a
modelentry inconfig.toml
Cloud chat default model selection cannot currently be changed. If cloud chat points to a retired model for your route, use a supported model on a different surface or sign in with an API key for local tasks where appropriate. See Codex Models for the authoritative list.
A Safe Triage Checklist Before You Change Settings
Before editing configuration files or switching authentication, work through a short, reversible checklist. Each step validates availability and clears common local pitfalls without risking data or credentials.
- Confirm your sign-in method and surface. Record whether you are on ChatGPT sign-in or API key, and whether the error occurs in the desktop app, CLI, IDE, or cloud chat.
- Check the official model list for your route. Read the retirement notes and recommended replacements on Codex Models.
- Run basic CLI diagnostics if you use local tools. In the Codex CLI, use
/statusand/modelto see your current state. Reference the Codex CLI docs. - Consult the Changelog for recent fixes affecting model switching or authentication.
| Step | Command or Check | Expected Outcome |
|---|---|---|
| Diagnostics | Run /status | Shows connectivity, auth route, and recent notices |
| Model check | Run /model | Prints the effective model value from config or session |
| Permissions | Run /permissions | Lists capabilities allowed by your route or workspace |
| Version awareness | View Codex CLI About | Note your installed version and compare to the Changelog |
These steps do not alter your files or tokens. They simply provide evidence to guide the next action.
Check the Surface Model and Workspace Context
Codex is available across multiple surfaces, and each surface can read a different source of truth for the model setting. The desktop app, Codex CLI, and IDE extension share a common config.toml, where you can set a model entry. A saved value of gpt-5.4 in that file will continue to drive local behavior even if you switched models in a different window. The cloud chat default model cannot currently be changed, so verify that the surface you are using supports the model you expect.
In managed workspaces, a policy can enforce a set of allowed models. If your personal profile can access a model but your team workspace cannot, the same command or task can behave differently. Collect a screenshot or a note of the workspace policy if you need to escalate to an admin.
If you are testing context heavy prompts while switching models, consider cost and performance trade offs in addition to availability. Our Codex long-context cost guide explains how long prompts can double billing at high token counts. For model pricing background, see the GPT-5.3-Codex pricing guide.
Verify Your CLI App and Configuration
The Codex CLI is a practical way to validate your local environment. On first run, you choose a sign-in method and the CLI will cache your login for reuse in compatible tools. The CLI supports commands such as /status, /permissions, /model, and workflows like codex exec to run tasks. For reference, review the official Codex CLI documentation.
Key checks:
- Inspect your effective model. Run
/modelto see the current value. If it shows gpt-5.4 and you are on ChatGPT sign-in after August 31, 2026, update your configuration to use a supported replacement. - Locate your configuration file. The desktop app, CLI, and IDE extension share a single
config.toml. Ensure themodelentry is set to a supported model. You can confirm paths and usage in the CLI docs. - Stay current, but do not chase a single version number. The Changelog documents fixes for model switching, onboarding, and reconnection. Update to the latest CLI available for your platform instead of relying on a specific historical version. Track changes on the Changelog.
If you want to compare frameworks or models for your workflow, consider the ecosystem. Our Codex versus Claude Code comparison and the Claude 4.7 agents guide outline differences in capability and ergonomics.
Refresh Cached Authentication Without Losing Credentials
Codex caches authentication to streamline local work. If permissions or entitlements changed since your last login, a refresh can clear stale state without deleting files.
Safe refresh steps:
- Logout in the CLI or desktop app using the built in sign-out action.
- Close local Codex tools that share cached credentials.
- Reopen the CLI or desktop app and sign in again, choosing either ChatGPT sign-in or API key as appropriate for your plan and task.
Where are credentials stored? Depending on your platform, Codex can store them in the OS credential store or in ~/.codex/auth.json. Treat this file like a password. Do not open it in screenshots, do not commit it to a repository, and do not share it. You do not need to delete it to resolve most sign-in related model errors.
For visibility into recent improvements to onboarding and reconnection flows, check the release notes in the Changelog. If you work in public repositories, you can also watch issue discussions in the OpenAI organization on GitHub for context on user reported behavior.
Choose a Supported Replacement Model
If you are on ChatGPT sign-in and request gpt-5.4 after its retirement date, switch to a supported model. The Codex Models page currently recommends three GPT-5.6 variants for different scenarios:
- Sol for complex, multi step work
- Terra as an everyday workhorse
- Luna for fast, repeatable tasks
In practice, replace gpt-5.4 with gpt-5.6-terra for general development. If your workload favors speed and repeated short actions, replace gpt-5.4-mini with gpt-5.6-luna. For demanding refactors or architecture changes, consider Sol when allowed by your workspace policy and budget. See Codex Models for current guidance.
| Task Type | Recommended Model | Why |
|---|---|---|
| General coding and review | gpt-5.6-terra | Balanced capability and latency for daily work |
| Short, repetitive tasks | gpt-5.6-luna | Optimized for speed and repeatability |
| Complex refactor or multi file change | GPT-5.6 Sol | Better for deep reasoning and larger changes |
| Cost sensitive, long context prompts | gpt-5.6-terra or luna | Manage token usage and latency while maintaining coverage |
If you design multi agent systems, you may mix models by role to balance cost and throughput. Our multi-agent coding architecture guide covers patterns for coordination, and our AI coding agent cost analysis highlights hidden burn scenarios to avoid.
When API Key Authentication Is Appropriate
The retirement of GPT-5.4 and GPT-5.4 mini applies to Codex when you sign in with ChatGPT. The OpenAI API and Codex sessions authenticated with an API key are not affected by that specific retirement. If your work is local and you need a model that remains available through the API, you can sign in with an API key in the desktop app, CLI, or IDE extension.
Considerations:
- Billing follows API rates when you use an API key. Track usage accordingly.
- Some ChatGPT workspace or Codex cloud features can be limited or unavailable with API key sign-in.
- Never paste or publish your API key. Prefer environment variables or your OS credential store, and keep
~/.codex/auth.jsonprivate.
If your project relies on retrieval augmented generation, plan model and token budgets with data flows in mind. For a primer on retrieval patterns and trade offs, read our RAG systems guide.
What to Do When the Error Persists
If you still see a not supported message after switching to a recommended model and refreshing authentication, collect evidence for support or your workspace admin. Include your sign-in method, surface, the exact model string, and timestamps. Confirm you followed the current docs and checked the Changelog for any relevant fix notes.
Additional actions:
- Re run
/statusto verify connectivity and any posted notices in the CLI. - Review the Codex Models list to ensure your new model is clearly supported on your route.
- Confirm your workspace policy still allows the replacement model. Ask an admin if in doubt.
- Test locally with API key sign-in when appropriate to isolate a workspace or ChatGPT sign-in specific constraint.
- Skim the Changelog to see if a recent update affects model switching or onboarding in your tool.
For broader planning that spans multiple agents and services, our multi-agent coding architecture guide discusses how to route tasks to the right capability tier without locking into a single model version.
Final Fix Order for gpt-5.4 Not Supported
Use this final, reversible sequence to resolve the issue without risky changes:
- Identify your route and surface. Note whether you are on ChatGPT sign-in or API key, and where the error occurs.
- Check the official model list. Read Codex Models for current availability and retirement status.
- Run local diagnostics. In the CLI, use
/status,/permissions, and/modelto confirm connectivity, entitlements, and the effective model value. - Replace retired models in config. Edit
config.tomlto setmodel = "gpt-5.6-terra"or another supported choice as documented. Save and relaunch. - Refresh cached authentication. Log out and sign in again. Do not delete
~/.codex/auth.json. Treat it like a password file. - Update your tool. Install the latest Codex CLI or desktop app. Use the CLI docs and compare behavior with notes in the Changelog.
- Escalate with evidence. If the error continues, collect your route, surface, model string, timestamps, and workspace policy for support or your admin.
Verdict: the error usually reflects a valid lifecycle or entitlement boundary rather than a universal bug. By confirming your route, switching to a supported model, and refreshing authentication, you can clear the block while preserving security and configuration hygiene.
Frequently Asked Questions
SK Jabedul Haque
Building India's most trusted finance education platform — simplifying news, schemes and market trends so anyone can understand and invest confidently.
Read full bioNever miss an update
Get our clearest explainers on schemes, markets and money — read what matters, without the noise.
Explore more articles