Short answer: OpenAI says o3 will be retired from ChatGPT on August 26, 2026 after a 90-day sunset period. That does not mean the o3 API is being shut down on the same date. OpenAI explicitly says this retirement applies to ChatGPT and that there is no API change from this announcement.
For people who still choose o3 manually, the practical question is therefore not “Will ChatGPT stop working?” It is: which newer reasoning option should replace the parts of a workflow where o3’s behaviour mattered?
This explainer was checked against OpenAI’s current public documentation on August 11, 2026.
What changes on August 26
OpenAI’s model release notes say o3 will be retired from ChatGPT on August 26. Once a model is retired from ChatGPT, it is no longer available as a selectable ChatGPT model.
That matters most to people who deliberately pick o3 rather than relying on ChatGPT’s default model selection.
OpenAI’s broader guidance on model transitions says prior conversations can carry over when the underlying model changes. Its legacy-model documentation also notes that retired models disappear from the picker and that existing conversations may move to a newer supported model.
So an old conversation does not necessarily disappear, but the fact that a chat originally used o3 should not be treated as a guarantee that future messages in that thread will continue running on o3 after retirement.
What does not change from this announcement
The most important distinction is between ChatGPT and the OpenAI API.
OpenAI’s retirement notice says the August 26 change applies to ChatGPT only. It explicitly states that there are no API changes associated with this retirement.
That gives three different cases:
| Current use | August 26 effect | What to do |
|---|---|---|
| Manually choosing o3 in ChatGPT | o3 is scheduled to disappear from ChatGPT | Test a current reasoning option before the deadline |
| Continuing an old o3 ChatGPT conversation | Conversation history can remain, but the underlying model may move | Re-check important outputs after the transition |
| Calling o3 through the OpenAI API | No change from this ChatGPT retirement announcement | Keep watching the API deprecation documentation separately |
The last row is easy to miss. A ChatGPT model retirement and an API model retirement are separate lifecycle events.
What replaces o3 in normal ChatGPT use?
There is no single universal “o3 replacement” because current ChatGPT exposes reasoning through newer models and different reasoning levels.
As of August 11, OpenAI says GPT-5.6 Sol is rolling out to eligible paid plans. In standard ChatGPT conversations:
- Medium and High reasoning use GPT-5.6 Sol on Plus, Pro, Business and Enterprise.
- Extra High is available on Pro, Business and Enterprise.
- Pro uses GPT-5.6 Sol Pro where that option is included.
- GPT-5.5 Instant remains the default for fast everyday responses.
- Free and Go do not currently include GPT-5.6 Sol in standard ChatGPT conversations.
This is why simply asking “Which model is the new o3?” is less useful than asking how much reasoning the task actually needs.
A practical replacement framework
| Workload | First option to test | Why |
|---|---|---|
| Everyday questions, rewrites, quick summaries | Instant | Using maximum reasoning for routine work usually adds latency without much value |
| Coding bugs, multi-step planning, careful analysis | Medium | Good first comparison point for former o3 workflows |
| Difficult debugging, research synthesis, complex constraints | High | More reasoning when Medium misses important interactions |
| Especially difficult or long-running work | Extra High or Pro, when available | Best reserved for tasks where additional reasoning materially changes the result |
This is not a benchmark claim that one setting will always beat o3. It is a migration strategy: start with the least expensive reasoning level that reliably passes the task’s acceptance criteria, then increase reasoning only when the lower level fails.
Do Plus or Pro users need to upgrade because o3 is retiring?
Not automatically.
OpenAI’s current GPT-5.6 availability table says Plus includes GPT-5.6 Sol at Medium and High reasoning. That means a Plus subscriber does not have to move to Pro merely because o3 disappears from ChatGPT.
Pricing checked August 11:
- Plus: $20/month according to OpenAI’s current Plus help page.
- Pro $100: 5× the usage allowance of Plus, according to OpenAI’s current Pro tiers page.
- Pro $200: 20× the usage allowance of Plus and remains the highest-usage Pro tier.
Both Pro tiers have the same core Pro capabilities; OpenAI says the main difference is usage allowance.
The useful buying question is therefore not “Which plan has the old model?” It is whether the workflow actually needs Extra High/Pro reasoning or substantially more usage than Plus provides.
Enterprise and Edu: legacy access is a transition tool, not permanent storage
OpenAI allows Enterprise and Edu workspace admins to enable legacy models while those models remain available. The current legacy-model documentation lists o3 among models that can appear when legacy access is enabled.
But the same documentation makes the lifecycle rule clear: legacy models can be removed as their retirement windows close, and retired models no longer appear in the ChatGPT picker.
For teams, that means “enable legacy models” is useful for migration testing, not as a strategy for avoiding a published retirement indefinitely.
A sensible team transition is:
- Identify prompts, GPTs or internal playbooks that explicitly depend on o3.
- Run representative examples on the likely replacement model.
- Compare outputs using a fixed rubric rather than subjective memory.
- Update documentation to describe the required behaviour, not the old model name.
- Keep API dependencies separate from ChatGPT dependencies because their retirement schedules are not the same.
The biggest migration mistake: testing with one impressive prompt
A model transition can feel better or worse based on a single conversation, but that is a weak test.
A more reliable approach is to build a small regression set from real tasks. Five to ten examples is already more informative than one carefully chosen demo.
For each task, record what success actually means. For example:
| Task | Acceptance test |
|---|---|
| Debug a production-like error | Finds root cause, proposes a valid fix, does not invent APIs |
| Compare two contracts or policies | Identifies every material difference and cites the correct section |
| Research a technical decision | Uses current sources and separates facts from uncertainty |
| Refactor code | Preserves behaviour and passes tests |
| Long analytical prompt | Follows all constraints without silently dropping later instructions |
Run the same set through o3 before August 26, then through the newer reasoning option being considered. Keep the prompt and scoring rubric fixed.
This produces a much more useful answer than “the new model feels smarter.”
A 20-minute o3 migration checklist
1. Find model-specific dependencies
Search saved prompts, Custom GPT instructions, team documentation and automation code for explicit references to o3.
Do not assume every mention is a real dependency. Some prompts name a model even though the task itself works fine on newer reasoning models.
2. Pick three representative tasks
Choose one routine task, one difficult task and one failure-prone task. Avoid cherry-picking only examples where o3 performs unusually well.
3. Run the same inputs on the likely replacement
Start with GPT-5.6 Medium for normal reasoning work, then try High if important errors remain. Use Extra High or Pro only when the added capability is actually useful and available on the plan.
4. Compare behaviour, not writing style
Score factual accuracy, constraint-following, completeness, tool use, code correctness or whatever matters to the workflow.
A different tone is not automatically a regression. A prettier answer is not automatically an improvement.
5. Separate ChatGPT from API dependencies
If an application calls o3 through the API, do not remove or rewrite that integration solely because of the August 26 ChatGPT deadline. OpenAI says the ChatGPT retirement does not change API availability.
Instead, track API deprecation notices independently and plan a separate migration when OpenAI publishes one.
What to watch next
Three things matter between now and August 26.
First, model-picker availability can still vary during rollout. OpenAI says GPT-5.6 is rolling out gradually, so an eligible account may not show every option immediately.
Second, managed workspaces can have administrator-controlled model access. A model being generally available does not guarantee that every Business or Enterprise member can select it.
Third, OpenAI’s current lifecycle pattern shows that model availability in ChatGPT changes faster than API availability. Workflows that rely on a model name should therefore treat the model as a replaceable implementation detail and keep a small acceptance-test suite ready for future transitions.
Bottom line
o3 is scheduled to leave ChatGPT on August 26, 2026, but this announcement does not retire o3 from the OpenAI API. Existing conversation history can remain, while future turns may move to a supported model once o3 is no longer available.
For most paid ChatGPT users, the sensible migration is to test the same real tasks on the current GPT-5.6 reasoning options before the deadline rather than waiting for the picker to change. Plus already includes Medium and High reasoning; Pro mainly adds higher reasoning options and substantially larger usage allowances.
The durable lesson is bigger than o3: save the behaviour you need as tests and rubrics, not as loyalty to a model name. Models will keep changing. A workflow with clear acceptance criteria is much easier to migrate than one built around “this prompt only works on model X.”
Sources
- OpenAI — ChatGPT release notes: o3 retirement on August 26, 2026
- OpenAI — Model release notes
- OpenAI — GPT-5.6 in ChatGPT: availability by plan and reasoning level
- OpenAI — What to expect when models change
- OpenAI — Legacy model access for Enterprise and Edu
- OpenAI — ChatGPT Plus
- OpenAI — About ChatGPT Pro tiers