What is GPT-5.5 Instant, ChatGPT's new default model?

Yesterday, Tuesday, May 5, 2026, OpenAI changed the model that answers when you open ChatGPT and simply start typing. The new default is GPT-5.5 Instant, and it takes over from GPT-5.3 Instant for every user, free or paid. If you use ChatGPT without picking a model, this is the one you are now talking to.

OpenAI says hundreds of millions of people rely on Instant every day, which is why a quiet update to it matters more to most users than a headline launch. The company focuses on three things: answers that are more accurate, answers that are shorter and clearer, and better use of what ChatGPT already knows about you, with a new way to see and correct that knowledge.

Instant, Thinking and the default

ChatGPT runs a family of models. An Instant model replies right away and handles most everyday questions. A Thinking model, such as GPT-5.5 Thinking released two weeks ago, takes more time to reason step by step before it answers and suits harder work. The default model is simply the one ChatGPT uses unless you choose otherwise.

Here is what changes and when:

The three-month overlap is worth noticing. When OpenAI removed GPT-4o from ChatGPT in February, many users protested because they had grown attached to that model's personality, as TechCrunch recalls. Keeping the older Instant model available for a while gives paying users time to adjust.

Fewer made-up facts

A hallucination is when an AI model states something false as if it were true, such as a wrong dosage, a law that does not exist or an invented statistic. It happens because a language model predicts likely-sounding text; it does not look facts up unless it uses a tool such as web search.

OpenAI reports two results from its own testing:

• On difficult prompts about medicine, law and finance, the false claims dropped by 52.5% compared with GPT-5.3 Instant.
• On tough conversations that users had flagged for factual mistakes, it made 37.3% fewer inaccurate claims.

It also scores higher on standard tests. TechCrunch reports 81.2 on AIME 2025, a hard high-school math competition, against 65.4 for the older model, and 76 against 69.2 on MMMU-Pro, a test of reasoning about images and text together. OpenAI says the model is better at reading photos you upload, at science and math questions, and at judging when to search the web instead of answering from memory.

These are OpenAI's own numbers. Fewer errors is not the same as no errors, so anything that affects your health, money or legal situation still needs checking with a reliable source or a professional.

Shorter, cleaner answers

The second change is style. OpenAI says GPT-5.5 Instant gets to the point faster while keeping a warm tone. It uses less heavy formatting, asks fewer unnecessary follow-up questions and drops decorative emojis. In one example OpenAI published, about advising someone on a talkative coworker, the new model's reply used about 30% fewer words than the old one while still offering ready-to-use phrases.

For most people, that means less scrolling and less skimming past headings and bullet points to find the answer.

Personalization, and seeing what it is based on

Personalization means ChatGPT shapes its answer around you: your earlier conversations, your saved memories, files you have shared, and Gmail if you connect it. GPT-5.5 Instant is better at deciding when this context would help and faster at finding the right past chat, so you repeat yourself less.

To make that visible, OpenAI is adding memory sources to all ChatGPT models. When a reply has been personalized, you can see what it drew on, such as a saved memory or an earlier chat, and remove or fix anything that is out of date. Memory sources stay hidden if you share the chat with someone else. OpenAI notes one limit: the list may show the most relevant items rather than every factor that shaped the answer.

Why it matters

Most people never touch the model menu, so the default model is effectively ChatGPT for them. An improvement in accuracy on medical, legal and financial questions reaches a very large number of people at once, including those least likely to double-check.

The personalization changes point in a clear direction: ChatGPT is becoming an assistant that knows your history. Memory sources are OpenAI's answer to the obvious question that raises, namely what it knows about you and where that came from.

How can you check and control what ChatGPT remembers about you?

Start with the answer itself. When a reply is personalized, look for its memory sources to see which saved memories or past chats it used. If something is wrong or old, such as a job you have left or a diet you no longer follow, delete or correct it right there.

For a wider cleanup, open the memory section of ChatGPT's settings, where you can review, edit or delete saved memories. You can also delete past chats you no longer want ChatGPT to draw on. When you want a conversation that leaves no trace in your memory, use a temporary chat, which neither reads nor updates it.

You can also simply tell it what to keep or forget:

Please forget that I live in Toronto; I moved to Ottawa.
Also remember that I prefer short answers with one example.
What do you currently remember about my preferences?

Then check the result in settings, since the list in your settings is the record that counts.