OpenAI’s August 7, 2025 GPT-5 rollout initially removed GPT-4o, o3 and other selectable models from many ChatGPT accounts. That decision was partly reversed within days, then followed by a permanent ChatGPT retirement program. As of August 18, 2026, GPT-4o is no longer available in ChatGPT, o3 is scheduled to disappear from ChatGPT on August 26, and OpenAI’s documentation says both remain available through the API. ChatGPT access and API access are separate questions.
What happened in August 2025
When GPT-5 became ChatGPT’s default on August 7, 2025, OpenAI initially removed older selectable models from the ordinary ChatGPT model picker. The models named in contemporaneous coverage included GPT-4o, o3, o4-mini and o4-mini-high. Enterprise and Edu customers were told they would have a limited legacy-access period, while developers were told that the older API models were not being deprecated at that point. Contemporaneous reporting described the resulting backlash.
Why the change disrupted people
- Users had tuned workflows to a model’s particular speed, tone, verbosity and refusal behavior.
- GPT-4o was popular for quick, conversational tasks, while o3 was designed for harder, multi-step reasoning and often felt less casual.
- Custom GPTs and saved chats could produce different results when their underlying model changed.
- Model choice let users trade off speed, creativity, reasoning depth and, in API settings, cost.
Reported reactions included objections to how GPT-4o’s retirement was presented, including a demonstration in which the model was asked to write a “eulogy.” That was a reported user reaction, not evidence that every user opposed GPT-5. OpenAI later said most usage had moved to newer models, while a smaller group still preferred GPT-4o’s style and capabilities. OpenAI’s retirement announcement provides that rationale.
The first removal was not permanent
OpenAI acknowledged rollout problems on August 8, 2025. On August 12 it restored GPT-4o to the model picker for paid users and added a “Show additional models” setting that exposed models including o3, o4-mini and GPT-4.1. The ChatGPT release notes record that reversal.
#1 Best Overall
Therefore, an account that could not select GPT-4o on August 7, 2025 was not necessarily experiencing a permanent shutdown. The later retirement dates, rather than the initial rollout, determine current availability.
The definitive ChatGPT retirement timeline
| Date | Event | What it means |
|---|---|---|
| April 16, 2025 | OpenAI introduced o3 and o4-mini in ChatGPT and the API. | o3 was both a ChatGPT and developer-facing model. |
| August 7, 2025 | GPT-5 rollout began; older ChatGPT models were initially removed for many users. | This is the event described by the original headline. |
| August 12, 2025 | GPT-4o returned for paid users; additional models appeared behind a setting. | The first removal was reversed. |
| February 13, 2026 | GPT-4o, GPT-4.1, GPT-4.1 mini, o4-mini and certain GPT-5 models retired from ChatGPT. | The main permanent ChatGPT retirement date. |
| March 11, 2026 | GPT-5.1 Instant, Thinking and Pro retired from ChatGPT. | The broader retirement program continued. |
| April 3, 2026 | GPT-4o left Business, Enterprise and Edu Custom GPTs. | The temporary workspace exception ended. |
| June 27, 2026 | GPT-4.5 retired from ChatGPT. | Another legacy model left the product. |
| August 26, 2026 | o3 is scheduled to retire from ChatGPT. | On August 18, o3 remains in a final transition window for eligible users. |
The dates above are documented in OpenAI’s o3 announcement, release notes, retirement guidance and model-release notes.
GPT-4o and o3 do not have the same status
GPT-4o
GPT-4o is retired from ChatGPT on every plan. Business, Enterprise and Edu customers had a temporary exception for GPT-4o inside Custom GPTs, but that ended April 3, 2026.
Rank #2
o3
o3 remains available in some Enterprise and Edu workspaces through the legacy-model system, and its stated ChatGPT retirement date is August 26, 2026. It was introduced as a reasoning model for difficult, multi-step work and remains a separate case from GPT-4o. The original announcement covers its launch.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesChatGPT access is not API access
| Question | ChatGPT | OpenAI API |
|---|---|---|
| Who controls availability? | OpenAI controls the model picker and can migrate chats or GPTs. | Developers select model IDs, subject to account permissions and API availability. |
| Best suited to | Individual use, collaboration and built-in tools. | Applications, agents, production integrations and controlled testing. |
| Retirement exposure | UI and workspace access can change quickly. | Models can still eventually be deprecated, but OpenAI says it will provide advance notice. |
| Migration control | Limited; users may be moved to a newer equivalent. | More control over model IDs, prompts, fallbacks and evaluation. |
| Billing | Subscription or workspace plan. | Usage-based API billing and related platform costs. |
OpenAI’s current documentation says models retired from ChatGPT continue to be available through the API and that future API retirements will receive advance notice. That is a current policy statement, not a promise of permanent availability. Access permissions, rate limits, pricing, latency and model behavior can still change. See the official retirement page, API platform and API documentation.
“Enterprise API remains” is therefore misleading. Enterprise is a ChatGPT workspace category; the API is a separate developer channel available according to account and model conditions.
Rank #3
What Enterprise and Edu users can still do
OpenAI provides a temporary legacy-model system for Enterprise and Edu administrators. Listed legacy models include GPT-4o, o3, o3 Pro and GPT-4.1. An administrator must enable the feature through workspace settings and the Models section; access is not automatic for every user. Legacy usage can also be subject to credits and limits. Details are in OpenAI’s legacy-model documentation.
This is a transition mechanism, not immunity from retirement. GPT-4o’s Custom GPT exception has already ended, and o3’s remaining ChatGPT access has a stated end date.
Free tools Windows power users keep installed
One-click scans. No signup required.
What happens to old chats and Custom GPTs
OpenAI says conversations and projects using retired models are moved to corresponding newer-model equivalents. GPTs using retired models are automatically moved to the closest GPT-5.3 Instant or GPT-5.4 Thinking/Pro equivalent. That migration can preserve access to the workflow while changing outputs, tone, tool calls or formatting, so important GPTs should be re-tested.
The text-model retirement does not automatically remove every feature associated with the “GPT-4o” name. OpenAI specifically says ChatGPT Voice is not changing as part of the text GPT-4o retirement, and ChatGPT Images is also not changing merely because the text model was retired. OpenAI’s current help page describes those distinctions.
A practical migration checklist
- Identify the event date. A missing model in August 2025 may have reflected the temporary rollout, while a missing GPT-4o model in 2026 reflects the retirement schedule.
- Check the plan and workspace. Free, Plus, Pro, Team, Business, Enterprise and Edu access has differed. Enterprise and Edu users should ask an administrator whether legacy models are enabled.
- Record API model IDs. For each application, document the exact ID, fallback behavior, limits and dependencies in the current OpenAI platform documentation.
- Build representative regression tests. Include real customer-support, coding, retrieval, structured-output and tool-use prompts rather than relying only on benchmark questions.
- Evaluate behavior, not just accuracy. Compare latency, token use, tone, refusal patterns, schema adherence, tool calls and downstream business outcomes.
- Preserve prompts and acceptance criteria. Saved examples make it possible to detect regressions after an automatic ChatGPT migration or an API model change.
- Plan a fallback. API applications can implement alternate model IDs or providers, but each fallback needs its own validation and data-governance review.
For teams that need side-by-side testing and monitoring, evaluation platforms such as LangSmith, Braintrust and Humanloop are examples of tools designed for model comparison and observability. They are most useful when an organization already has a regression dataset.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What this means for choosing a product
ChatGPT subscriptions
Paid ChatGPT plans provide newer models, higher limits and integrated features without API development. They are a poor fit when a business requires a permanently pinned legacy model: plan names, limits and model access can change. The product is at chatgpt.com.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Best Value
Business, Enterprise and Edu
These plans add administration and workspace controls, which can make a managed migration easier. They do not permanently preserve retired ChatGPT models. Enterprise information is available at OpenAI’s business page, with workspace model details at the model-limits documentation.
API and provider diversification
The API is the better fit for developers who need explicit model IDs, automated evaluations and fallback logic. Teams may also assess Claude, Gemini or the Gemini API, but should compare retirement policies, context limits, tool calling, data retention, regional availability, enterprise controls, pricing and rate limits before switching.
Frequently Asked Questions
Is GPT-4o still available in ChatGPT?
No. GPT-4o retired from ChatGPT on February 13, 2026; the temporary Business, Enterprise and Edu Custom GPT exception ended April 3, 2026.
Can I still use GPT-4o or o3 through the API?
OpenAI’s current documentation says models retired from ChatGPT remain available through the API, with advance notice for future API retirements. Availability and account limits can still change.
When does o3 leave ChatGPT?
OpenAI’s model-release notes schedule o3’s ChatGPT retirement for August 26, 2026. As of August 18, 2026, eligible Enterprise and Edu workspaces may still have access through legacy-model settings.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
