GPT-5.2 Is Gone from ChatGPT — What the New 90-Day Sunset Policy Means for You

As of June 12, 2026, GPT-5.2 Instant, GPT-5.2 Thinking, and GPT-5.2 Pro are no longer available in ChatGPT. If you had conversations running on any of those models, they have been automatically moved to the corresponding GPT-5.5 variant. No prompt, no confirmation dialog, no opt-out. It just happened.

For most people scrolling through ChatGPT on a Tuesday morning, this is invisible. For anyone with carefully tuned prompts, a workflow that depends on a specific model’s behavior, or a product built on top of a particular checkpoint, it is worth paying attention to.

What Actually Changed

GPT-5.2 launched on December 11, 2025, in three modes: Instant (fast, everyday tasks), Thinking (reasoning with optional extended thinking), and Pro (the highest-capability tier). It served as the workhorse for millions of users through early 2026 as OpenAI moved through a rapid release cycle, shipping GPT-5.3-Codex in February, GPT-5.4 in March, and GPT-5.5 in April.

The migration is model-for-model: GPT-5.2 Instant conversations now continue on GPT-5.5 Instant, Thinking maps to Thinking, and Pro maps to Pro. OpenAI announced the retirement alongside the release of GPT-5.3 Instant, so this date was not a surprise to anyone watching the release notes closely.

This is not the first time OpenAI has done this quietly. On March 11, 2026, GPT-5.1 Instant, Thinking, and Pro were retired from ChatGPT in exactly the same way, with conversations rolling forward to what was then the current model.

The 90-Day Sunset Policy

The more important thing to understand here is the rule OpenAI has now formalised: models will generally remain available in ChatGPT for 90 days after a successor is released. Once that window closes, they are gone from the interface and existing conversations migrate automatically.

GPT-5.5 was released on April 23, 2026. Count 90 days forward and you land squarely on the June 12 retirement date for GPT-5.2. The policy is working exactly as described.

Two other retirements are already on the calendar under this framework:

GPT-4.5 retires from ChatGPT on June 27, 2026, following a shorter 30-day sunset period (an explicit exception to the general 90-day rule, and only available to paid users in the meantime via model settings).
OpenAI o3 retires from ChatGPT on August 26, 2026, following the standard 90-day window. It will remain in the API for now.

If you are using either of those models in ChatGPT today, you have a clear deadline. The retirement date is not speculative.

What This Means for Your Prompts and Workflows

For casual users, the practical difference between GPT-5.2 and GPT-5.5 is unlikely to be noticeable in day-to-day use. GPT-5.5 Instant has been the default model for all logged-in users since May 5, 2026, so most people have already been using it without thinking about it.

The group who should pay attention is anyone who had deliberate reasons for being on GPT-5.2. Model behavior is not perfectly consistent across versions. Tone, verbosity, reasoning style, and instruction-following can all shift between checkpoints. A prompt that produced a reliable output on GPT-5.2 Thinking may behave differently on GPT-5.5 Thinking, even if the task is the same.

The right move is straightforward: go back and test your most important prompts on the new model. Do not assume the output will be identical.

For developers, the API picture is also relevant. On May 8, 2026, OpenAI notified developers using gpt-5.2-chat-latest and gpt-5.3-chat-latest model snapshots of their deprecation and removal. If you are still calling those aliases, you need to update your integration. OpenAI’s deprecations documentation is the place to track what is current.

What GPT-5.5 Actually Looks Like Now

GPT-5.5 is not just GPT-5.2 with a higher version number. There are meaningful changes worth knowing about.

GPT-5.5 Instant has been updated to improve response style: fewer bullet-heavy walls of text, better pacing in practical tasks, and a warmer conversational tone. If you previously found GPT-5.x responses overly structured or verbose, this is a deliberate change in the right direction.

GPT-5.5 Thinking handles harder tasks more efficiently, does a better job tracking where it is in a long reasoning chain, and now supports mid-task instructions. You can add a message while it is still thinking to redirect it. That is a genuinely useful addition for anything that requires back-and-forth on a complex problem.

One feature that has been removed from GPT-5.5 Instant and Thinking: Canvas. Writing and coding work is now handled directly in chat through writing blocks and code blocks rather than a separate canvas interface. Paid users can still access Canvas through legacy models until those models reach their own sunset dates.

There is also a new memory control worth knowing about: you can now delete specific memories shown on your memory summary page, and use a “Delete and turn off memory” option that clears saved context without deleting your chat history. If memory is turned back on later, ChatGPT may create new memories from existing conversations. This is rolling out across web and mobile for all ChatGPT users.

The Bigger Picture

OpenAI is releasing model updates quickly and retiring older ones on a predictable schedule. The 90-day sunset policy is a genuine commitment to that predictability, which is more than existed a year ago. Knowing the rule is useful: when a successor model is announced, start a 90-day countdown for whatever it replaces.

The thing that catches people out is not the retirement itself. It is the assumption that a forced migration is invisible. It mostly is, right up until the moment a workflow produces a different result and nobody can explain why. Checking your most-used prompts against the current model takes twenty minutes and removes that risk entirely.

GPT-5.2 had a good run. GPT-5.5 is the new baseline. The ChatGPT release notes are updated regularly and worth bookmarking if you rely on specific model behavior for anything that matters.