GPT-5.5 Instant becomes ChatGPT’s default to reduce hallucinations
- OpenAI made GPT-5.5 Instant the default ChatGPT model on May 5, replacing GPT-5.3 Instant for logged-in users to cut hallucinations without slowing replies. - OpenAI says GPT-5.5 Instant cut hallucinated claims 52.5% on high-stakes prompts and inaccurate claims 37.3% on user-flagged factual mistakes. - This matters because ChatGPT’s default model now auto-routes more carefully, making everyday answers safer without forcing everyone into slower reasoning.
ChatGPT’s default model just changed, and the point is pretty simple — make fewer things up without making the app feel slower. On May 5, OpenAI switched logged-in users from GPT-5.3 Instant to GPT-5.5 Instant, which now sits underneath the normal ChatGPT experience. The company is pitching it as a better everyday model, not a specialist one. That matters because the default model is the one most people actually use, especially when they are not thinking about model menus at all. (openai.com) ### What actually changed in ChatGPT? The main change is the default. GPT-5.5 Instant now replaces GPT-5.3 Instant for logged-in ChatGPT users, and OpenAI says the system is part of a single auto-switching experience that can decide when a faster answer is enough and when deeper reasoning is worth using. So the user sees one main assistant, but the stack underneath is getting more selective about how it answers. (hel([openai.com)3-gpt-55-in-chatgpt)) ### Why call it “Instant”? “Instant” is OpenAI’s fast-response tier — the model meant for everyday work, learning, how-tos, writing, and general questions. The whole selling point is low latency. You ask something, it answers quickly, and it does not always pause to do the heavier chain-of-thought style work that more deliberate models use. GPT-5.5 Instant is supposed to keep that feel while improving the quality of the answer itself. (openai.com) ### So how much better is it? The most important claim is about hallucinations. OpenAI says GPT-5.5 Instant produced 52.5% fewer hallucinated claims on high-stakes prompts and 37.3% fewer inaccurate claims on conversations users had flagged for factual errors. The company also says the model improved on image understanding, STEM questions, and web-search decisions. Those are internal numbers, so treat them as vendor b(openai.com)at OpenAI is optimizing for now — not just raw capability, but reliability in normal use. (openai.com) ### Why focus on hallucinations now? Because the default chatbot has a different problem than a frontier demo. Most people are not stress-testing abstract intelligence. They are asking for tax explanations, medication context, contract summaries, school help, and workplace drafting. In those settings, a fast wrong answer is worse than a slightly slower careful one. OpenAI and outside coverage both framed law, medicine(openai.com)ade is meant to land hardest. (techcrunch.com) ### What’s the trick under the hood? Basically, routing. OpenAI describes ChatGPT as a single experience that can auto-switch across models, using the fast workhorse by default and pulling in deeper reasoning when needed. That means the product direction is less “pick your model manually every time” and more “let the system decide how much thinking this question deserves.” The catch is trust — users need to believe the router knows when not to bluff. (help.openai.com) ### Is this also about style? Yes — and that is not cosmetic. OpenAI says GPT-5.5 Instant gives clearer, more concise answers and better personalization controls. Some coverage also noted fewer gratuitous emojis. That sounds minor, but concision matters when people use ChatGPT as a work tool. A shorter answer with fewer wrong side roads often feels smarter because, turns out, it usually is more usable. (openai.com) ### Why does this matter beyond one model swap? Because it shows where consumer AI is heading. The race is no longer just “who has the biggest model.” It is “who can make the default assistant dependable enough that ordinary users stop double-checking everything.” OpenAI is betting that the next big product win is invisible — fewer errors, cleaner answers, and smarter routing in the background. If that works, the model menu matters less, and trust matters more. (openai.com) ### Bottom line This is a product-quality update disguised as a model launch. GPT-5.5 Instant is now the ChatGPT most people get by default, and the whole point is to make everyday answers feel faster to trust, not just faster to generate. (openai.com)
