OpenAI has swapped the default ChatGPT model again, and this one matters because it is aimed at the part of ChatGPT people touch all day: the fast, everyday model that answers most prompts before anyone thinks about model names.
The new default is GPT-5.5 Instant. OpenAI says it is rolling out starting today, May 5, 2026, to all ChatGPT users, replacing GPT-5.3 Instant. The company also says the model is available in the API as chat-latest, which means developers using that moving alias should see the change without manually picking a new dated model.
What changed
The headline improvement is factuality. OpenAI says GPT-5.5 Instant produced 52.5 percent fewer hallucinated claims than GPT-5.3 Instant in internal high-stakes evaluations covering areas like medicine, law, and finance. It also says inaccurate claims dropped 37.3 percent on especially difficult conversations that users had previously flagged for factual errors.
That does not mean ChatGPT is suddenly a source of truth, and it definitely does not turn it into a doctor, lawyer, accountant, or security analyst. But it is still the right target. The default model does not need to win every benchmark nobody opens. It needs to stop inventing facts in the middle of normal work.
Less performative, more direct
OpenAI is also tuning the way the model talks. The company describes GPT-5.5 Instant as clearer and better at following user preferences, with fewer unwanted flourishes. One small but very visible example from the announcement is cutting down on gratuitous emojis.
That sounds cosmetic until you remember how many people use ChatGPT inside professional workflows. A model that gives direct feedback, preserves tone, and avoids theatrical filler is more useful than one that constantly tries to sound delighted.
Personalization is becoming the real product
The other important part of the update is personalization. OpenAI says GPT-5.5 Instant works better with ChatGPT's saved context, memory, custom instructions, and connected apps. The announcement frames this as a model that can use more of what ChatGPT knows about your preferences without overdoing the imitation.
This is where the product direction is obvious. ChatGPT is no longer just a box for one-off prompts. OpenAI wants the default experience to feel more like a work surface that remembers your style, understands recurring tasks, and can pull from the context you have allowed it to use.
That makes the controls more important. Memory and connected apps can make ChatGPT far more useful, but they also raise the cost of sloppy defaults. Users need to know what the model can see, what it remembers, and how to wipe or narrow that context when it gets something wrong.
Why this update matters
GPT-5.5 itself already rolled out in April for paid ChatGPT tiers and Codex. Today's change is different because it moves the newer behavior into the default lane. Most users do not live in model pickers. They open ChatGPT, type the thing, and expect the answer to be better than yesterday's answer.
If OpenAI's internal numbers hold up in normal use, GPT-5.5 Instant should make ChatGPT feel less slippery on factual questions, less noisy in tone, and more aware of user-specific preferences. Those are boring upgrades in the best way. They are the kind that reduce friction instead of asking users to learn a new workflow.
The rollout also makes chat-latest more interesting for developers who want the newest general ChatGPT behavior without pinning a model. That convenience comes with the usual tradeoff: if you need predictable behavior for production software, a moving alias is still something to test carefully before relying on it.
The bottom line
GPT-5.5 Instant is not being pitched as a flashy new frontier model. It is a cleanup pass on the model most people actually use: fewer hallucinations, better personalization, and less output that feels like it was written to impress a demo audience.
That is a good direction. The default ChatGPT model should be boringly reliable, easy to steer, and careful about what it claims. GPT-5.5 Instant is OpenAI's latest attempt to make that everyday experience feel less like a chatbot trick and more like a dependable tool.
