ChatGPT now defaults to GPT-5.5 Instant to reduce hallucinations

ChatGPT’s default brain just changed. OpenAI swapped in GPT-5.5 Instant on May 5, and the whole pitch is simple — keep the speed people expect, but make the answers less wrong. That sounds incremental, but for a product used constantly for homework, coding, planning, and quick facts, “less wrong” is the whole game. The update also tweaks how personalization works, so ChatGPT can show more clearly what stored context it used when answering. (openai.com) ### What actually changed? The old default model was GPT-5.3 Instant. The new default is GPT-5.5 Instant, and OpenAI says it is rolling out to all ChatGPT users, not just paid ones. Paid tiers still get more control through the model picker — Plus, Pro, and Business users can manually choose GPT-5.5 Instant or GPT-5.5 Thinking, while GPT-5.5 Pro is available on higher-end plans. (openai.com)ord? Because this is the model most people hit all day long. OpenAI’s bigger reasoning models can do more deliberate work, but they take longer and cost more attention. Instant is the daily-driver version — the one that has to answer fast enough that the product still feels conversational. So this release is really about pushing reliability into the quick model instead of asking users to trade speed for trust every time. (openai.com) ### How much less wrong is it? OpenAI put real numbers on it, which is the part that gives the launch weight. In its internal tests, GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in medicine, law, and finance. On especially difficult conversations that users had flagged for factual errors, inaccurate claims dropped 37.3%. Those are internal benchmarks, so take th(openai.com)much more concrete than the usual “smarter and better” language. (openai.com) ### What else changes in the chat itself? OpenAI says answers should feel tighter and more concise. One funny detail is that the company explicitly says the model uses fewer gratuitous emojis. But the more meaningful change is better handling of context — both what you just said in the chat and what ChatGPT already knows about your preferences when memory is on. The aim is to make responses feel more tailored without turning them into rambling, overconfident guesswork. (openai.com) ### What are the new memory controls? Basically, OpenAI is trying to make personalization less mysterious. The help documentation says users can manage memory, turn it off, delete saved details, and inspect what ChatGPT remembers. The GPT-5.5 materials add “memory sources” controls, which are meant to show which stored context shaped an answer. That matters because personalization is useful right up until it feels sp(openai.com)r. (openai.com) ### Why do hallucinations matter so much here? Because the default model is the one people trust by accident. A slower “thinking” model signals that you’re asking for something heavier. A default chat box does the opposite — it invites casual use, which means users often won’t double-check unless something looks obviously off. Reducing hallucinations in law, medicine, and finance is really about shrinking the danger (openai.com)lieved. (openai.com) ### Is this a big model launch or a product tuning move? More the second. GPT-5.5 itself launched in late April as OpenAI’s stronger flagship family, but this week’s move is about where those capabilities land inside ChatGPT day to day. In other words, OpenAI is taking improvements from the frontier-model stack and pushing a safer, cleaner version into the default consumer experience. That is less flashy than a brand(openai.com)ple actually use ChatGPT. (openai.com) ### Bottom line This is OpenAI admitting that the default chatbot experience lives or dies on trust. Speed still matters. Personality still matters. But the product only gets more useful if the fast answer is also the answer you don’t have to second-guess quite so much. (openai.com)