OpenAI has put two more legacy ChatGPT models on a fixed retirement schedule. In model release notes published May 28, 2026, the company said GPT-4.5 will leave ChatGPT on June 27, 2026 after a 30-day sunset period, while OpenAI o3 will leave ChatGPT on August 26, 2026 after a 90-day sunset period. OpenAI also said the affected models are available to paid users only through model settings, and that the change applies to ChatGPT only, with no API change.
That last point matters. This is not an API shutdown notice, and it is not a reason for developers to rewrite production API integrations overnight. It is, however, another sign that ChatGPT’s model picker has become a managed release channel with short transition windows. If a team has saved prompts, custom GPTs, training material, support playbooks, or creative workflows that depend on the feel of a particular ChatGPT model, the clock is now part of the product.
The dates are the story
The near-term date is GPT-4.5. A June 27 retirement gives teams less than a month from OpenAI’s May 28 notice to test replacements, update internal docs, and warn users who still select it manually. GPT-4.5 was introduced as a research preview for ChatGPT Pro users in February 2025, with OpenAI framing it as a large chat model built for natural interaction, writing, practical problem solving, and lower hallucination rates. Those are exactly the kinds of “feel” differences that tend to live in saved workflows rather than in benchmark dashboards.
OpenAI o3 gets a longer window, but it may be more operationally sensitive. OpenAI describes o3 as a strong reasoning model for coding, math, science, visual perception, and multi-step analysis. If a paid user or workspace deliberately routes hard questions, image analysis, spreadsheet reasoning, or coding tasks to o3, the August 26 date should be treated as a migration deadline rather than a preference toggle.
The difference between a 30-day and 90-day sunset also tells readers how OpenAI appears to be triaging old models. GPT-4.5 looks like a short-cycle retirement. o3 gets a longer runway, likely because reasoning workflows are more fragile and harder to replace invisibly. That is The AI Feed’s read, not an OpenAI claim, but the operational consequence is clear: start with the shorter GPT-4.5 deadline, then use the summer to test o3 alternatives before August 26.
ChatGPT and API calendars are splitting
OpenAI’s May 28 note is explicit that o3 and GPT-4.5 are being retired from ChatGPT, not from the API. That distinction is easy to miss because OpenAI also keeps a separate API deprecations calendar with its own deadlines. The API calendar includes unrelated 2026 shutdowns, including the Assistants API sunset on August 26, 2026 and later legacy model snapshot shutdowns. Same company, different surface, different migration work.
For builders, that means the first audit question is not “which OpenAI model do we use?” It is “where do we use it?” A ChatGPT workspace may lose a manually selected model while an API integration keeps running. An API migration may be urgent even if the ChatGPT UI still works. A team that blurs those surfaces will either overreact to a ChatGPT-only notice or miss a real API deadline.
The practical split looks like this:
- ChatGPT users should check saved chats, custom GPTs, workspace instructions, shared internal prompts, and training material that tells people to choose GPT-4.5 or o3.
- Workspace admins should check whether legacy model access is enabled and which teams still rely on it.
- Developers should separately review OpenAI’s API deprecations page instead of assuming ChatGPT sunset dates apply to production endpoints.
- Support and enablement teams should update screenshots and guidance before users discover the missing model in the picker.
Canvas is another clue
The same May 28 release notes also say OpenAI is updating GPT-5.5 Instant in ChatGPT and the API, and that canvas will no longer be available in GPT-5.5 Instant or GPT-5.5 Thinking. OpenAI says writing and coding work is now supported directly in chat responses through writing blocks and code blocks, while paid users can continue using canvas for a limited time through legacy models until those models are sunset.
That turns the model retirement into a feature retirement for some users. If a team still uses canvas because it likes the older editing surface, the relevant deadline is not only “which model disappears?” It is “which workflow disappears with it?” The risk is less about raw model quality and more about muscle memory: docs, demos, onboarding videos, and customer instructions may point to a UI path that is being phased out.
What to test before June 27
The safest migration plan is boring and specific. Pick a small set of representative tasks that GPT-4.5 or o3 handled well, run them on the current default model and the likely replacement reasoning model, and save the outputs side by side. Do not only test generic prompts. Test the awkward work: long files, visual inputs, code review, sensitive tone, customer-facing drafts, spreadsheet reasoning, and tasks where the old model had a known style advantage.
For GPT-4.5, the highest-risk workflows are likely writing, ideation, and tone-sensitive chat tasks. For o3, the highest-risk workflows are reasoning-heavy tasks where users intentionally selected a slower or more deliberate model. If the replacement is better on capability but different in behavior, update prompts and instructions around that difference instead of pretending nothing changed.
Teams should also capture a simple owner for each dependency. A custom GPT that names o3 in its instructions needs a maintainer. A sales enablement page that tells users to pick GPT-4.5 needs a date-stamped edit. A support macro that explains the model picker needs a new screenshot. None of that is glamorous, but it is the work that prevents “the model disappeared” from becoming a help-desk event.
What to watch next
The next checkpoint is June 27. If GPT-4.5 disappears on schedule, it will confirm that OpenAI is comfortable giving some paid ChatGPT legacy models only a 30-day off-ramp. After that, watch whether OpenAI publishes more replacement guidance for o3 before August 26, especially for users who rely on visual reasoning, coding, or longer analysis.
The broader pattern is already visible. OpenAI retired GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and older GPT-5-era ChatGPT models earlier in 2026 while saying the API was unchanged. Now o3 and GPT-4.5 are entering the same cycle. The model picker is not an archive. It is a moving product surface, and teams that treat it like infrastructure need their own lifecycle discipline.
For readers tracking OpenAI’s broader product and model strategy, see our OpenAI company profile and AI model leaderboard.
