Did the Codex Default Model Change to GPT-5.4-Codex-Mini?

If you searched for codex default model gpt-5.4-codex-mini changelog default model changed, you are probably trying to answer one of three questions:

1. Did Codex quietly switch its default model?
2. Is gpt-5.4-codex-mini the real model name?
3. What do I need to update before my local setup breaks or gets more expensive?

Here is the clean answer.

OpenAI's current Codex docs do not present gpt-5.4-codex-mini as the official model name. The documented model names are GPT-5.4 and GPT-5.4 mini. The same docs also say that GPT-5.4 and GPT-5.4 mini retire from Codex with ChatGPT sign-in on August 31, 2026, and that the replacement path is GPT-5.6 Terra for balanced daily work and GPT-5.6 Luna for cheaper, faster routine work.

So yes, there is a real default-model story here. But the more useful interpretation is not "Codex renamed everything to GPT-5.4-Codex-Mini." The useful interpretation is "my saved defaults, scripts, or team instructions may now point at an aging model family and need cleanup."

[stat] August 31, 2026 OpenAI's documented retirement date for GPT-5.4 and GPT-5.4 mini in Codex with ChatGPT sign-in

The Short Operational Answer

If you are signed into Codex with ChatGPT, treat the GPT-5.4 family as a winding-down model line and update your defaults now.

Question	Short answer
Is gpt-5.4-codex-mini the official documented name?	No
Are GPT-5.4 and GPT-5.4 mini real Codex models?	Yes
Do those models retire in Codex with ChatGPT sign-in?	Yes, on August 31, 2026
What should replace them in saved configs?	GPT-5.6 Terra or GPT-5.6 Luna
Does this apply to API-key workflows the same way?	No, OpenAI says the API-key path is not affected by this Codex retirement

That is the part most people actually need. Everything else is cleanup strategy.

Why This Query Exists At All

This search query looks like a mash-up of three separate ideas:

• people noticing a default model change
• people reading a Codex changelog
• people compressing GPT-5.4 mini into an unofficial "codex-mini" style name

That naming confusion is normal. Codex has had several model families, several naming patterns, and different behavior depending on whether you are using:

• Codex with ChatGPT sign-in
• Codex with your own API key
• local defaults in config.toml
• saved workspace or automation settings

Once teams start copying model strings into scripts, README files, and onboarding docs, the wrong name tends to spread faster than the right one. Search demand then follows the mistake.

That is exactly why this page exists.

What The Changelog Actually Means For Teams

The mistake most teams make is thinking a model-default change is just cosmetic. It is not.

A default-model change affects:

• new Codex sessions started by teammates who never set an explicit model
• old config files that still pin yesterday's favorite model
• scheduled tasks and automations that quietly inherit outdated defaults
• team expectations around speed, quality, and usage burn

The real risk is not only "my command breaks." The real risk is silent drift:

• an agent loop gets slower because it is still pinned to the wrong model
• a routine task uses a stronger model than necessary
• internal docs keep telling people to use a model that is about to retire

If you want to keep costs under control, model hygiene matters just as much as prompt hygiene.

---

What To Update Right Now

If you manage a Codex setup for yourself or a team, check these surfaces first:

1. config.toml
2. repo-level defaults or managed config
3. custom agents and saved model presets
4. scheduled tasks
5. onboarding docs, internal runbooks, and shell aliases

Use this rule of thumb:

• Choose GPT-5.6 Terra when you want the balanced everyday default.
• Choose GPT-5.6 Luna when the work is narrower, faster, and more repetitive.
• Review any lingering GPT-5.4 mini references before they turn into support noise later.

If your team also runs API-key workflows, separate those decisions from ChatGPT-sign-in Codex decisions. OpenAI explicitly distinguishes the two, and mixing them creates bad migration advice.

A simple migration checklist

Use this when cleaning up a Codex setup:

1. Search for gpt-5.4, gpt-5.4-mini, gpt-5.3-codex, codex-mini, and model =.
2. Check config.toml, workspace defaults, and automation configs first.
3. Review team docs and onboarding steps second. Bad model strings spread through copy-paste faster than code review catches them.
4. Audit scheduled tasks and wrapper scripts third. They are where stale defaults usually survive the longest.
5. Decide which flows should move to GPT-5.6 Terra and which should move to GPT-5.6 Luna.
6. Re-run one real task after the swap so you catch accidental performance or cost regressions immediately.

That last step is the part most teams skip. It is also the step that tells you whether the migration was actually successful.

Where Default-Model Drift Usually Hides

Most teams imagine a default-model change as a developer laptop problem. It is usually an automation problem instead.

Repo-level config and agent instructions

This is the obvious source, but it is still worth naming. A repo can pin a model in config, in a setup script, or in written instructions that new teammates blindly follow.

Scheduled tasks

Background jobs are perfect places for outdated model names to survive because nobody stares at them every day. Check:

• scheduled code review jobs
• maintenance scripts
• cron-based content tasks
• shell aliases for codex exec
• helper scripts inside internal tooling

Internal documentation

One engineer posts a command in chat, someone copies it to the wiki, and three weeks later the whole team assumes it is canonical. That is how unofficial strings like gpt-5.4-codex-mini get real search demand.

Mixed authentication modes

This is the most annoying case. Some teams use ChatGPT sign-in, others use API keys, and the model-availability rules are not always the same. If your internal docs blur those paths together, you get migration advice that is half right and fully useless.

[vs] ChatGPT sign-in Codex || API-key Codex Same product family, different model-availability and retirement rules. Treat them as separate operational surfaces.

The Cost-Control Angle Most Teams Miss

AI Cost Check is about cost, so here is the blunt version.

The worst default model is not the "wrong" model in some abstract purity sense. The worst default model is the one you forget to revisit.

That is how small inefficiencies become recurring spend:

• a stronger model becomes the fallback for routine edits
• a retired model string lingers in scripts until a failure forces cleanup
• nobody documents when to use the cheaper fast path

The better pattern is explicit routing:

• premium model for ambiguous, high-stakes work
• balanced model for normal daily work
• cheaper model for repetitive tasks, extraction, summaries, and cleanup

If you are using Codex operationally, that routing logic is more valuable than obsessing over one changelog screenshot.

For API planning, compare likely workloads in the AI cost estimator, then review nearby model pages such as GPT-5.6 Terra, GPT-5.6 Luna, and GPT-5.4 mini. That gives you a more durable plan than chasing rumors about the default.

How to choose Terra versus Luna

Do not replace everything with the same newer model just because it feels tidy. Tidy is not the same as cost-aware.

Use GPT-5.6 Terra when you need:

• stronger day-to-day reasoning
• better synthesis across tools or multiple files
• a safe default for mixed work where quality matters
• fewer surprises on ambiguous tasks

Use GPT-5.6 Luna when you need:

• faster turnaround
• narrowly scoped tasks
• repetitive cleanups
• a cheaper default for high-volume routine work

If the task has a crisp definition of done, Luna is usually the smarter default. If the task is open-ended or higher stakes, Terra is safer.

Why This Query Is Valuable SEO Traffic

This is not a giant vanity keyword. It is a narrow operational query with strong intent. People who search it usually want to:

• confirm whether the default changed
• figure out the correct model name
• update a config before something breaks
• understand whether the change affects API-key workflows too

That is why a sharply targeted explainer can win clicks here even when broad keywords like "AI pricing" or "API cost calculator" are still buried deep in the rankings. The intent is tighter, and the answer is binary. Either the result clearly resolves the confusion, or it does not.

The current zero-click pattern makes sense. Searchers probably saw results adjacent to the topic but not explicit enough to answer the naming confusion. A page that directly says "no, that exact string is not the documented name, but yes, you should update your defaults" is far more clickable.

What Good Team Documentation Looks Like After This Change

If you manage Codex for a team, document the migration like this:

For ChatGPT-sign-in Codex

• state the everyday default model
• state the cheaper routine-work model
• note the retirement date for older model families
• link one source-of-truth migration note instead of repeating advice everywhere

For API-key workflows

• document those model choices separately
• avoid assuming ChatGPT-sign-in retirement notes apply automatically
• keep scripts, runbooks, and examples aligned with the same source of truth

For automations

• pin the model explicitly when cost or consistency matters
• review model references on a schedule
• avoid vague language like "use the default model" in cron jobs and scheduled agents

That last point sounds boring. Good. Boring documentation is what keeps operations cheap.

Recommended Team Policy

If you run Codex across multiple repos, use this simple policy:

1. Stop relying on hidden defaults

Document the intended model per workflow. If a task is cost-sensitive, say so explicitly.

2. Split everyday work from heavy work

Use one default for normal throughput and a second model for complex jobs. That prevents expensive models from becoming accidental daily drivers.

3. Audit scheduled tasks and agent configs monthly

This is where stale model strings survive the longest.

4. Treat changelog-driven model changes as an ops task

When a model family is being retired, assign cleanup instead of assuming everyone will notice.

That is boring advice, which is exactly why it works.

Frequently asked questions

Is gpt-5.4-codex-mini the official model name?

No. The current Codex docs describe GPT-5.4 and GPT-5.4 mini as the documented names. The longer phrase looks like community shorthand or a naming mash-up.

Did Codex defaults actually change?

The practical answer is yes. OpenAI's current Codex docs recommend the GPT-5.6 family for Codex with ChatGPT sign-in and document retirement for GPT-5.4 and GPT-5.4 mini on August 31, 2026.

Should I replace everything with GPT-5.6 Terra?

No. Use GPT-5.6 Terra for balanced everyday work and GPT-5.6 Luna for clearer, cheaper routine work. Blanket upgrades are lazy and usually waste money.

Does this affect API-key workflows too?

OpenAI's current Codex docs say the GPT-5.4 retirement in Codex does not affect API-key workflows in the same way. That is why your migration note should separate those two paths explicitly.

Why does this matter for an AI cost site?

Because it is high-intent traffic. People searching this query are already in ops mode. If you answer the confusion clearly, you can win the click without needing massive impression volume.

Final Answer

If you came here because you saw the query codex default model gpt-5.4-codex-mini changelog default model changed, the safest conclusion is:

• the exact phrase gpt-5.4-codex-mini is not the official current Codex model name in OpenAI's docs
• the real documented models are GPT-5.4 and GPT-5.4 mini
• OpenAI says those models retire from Codex with ChatGPT sign-in on August 31, 2026
• the documented replacement direction is GPT-5.6 Terra and GPT-5.6 Luna

So do not waste time debating the nickname. Update the config, clean the docs, and move on.