ChatGPT now defaults to GPT‑5.5 Instant to curb hallucinations on high‑stakes prompts
- OpenAI began rolling out GPT‑5.5 Instant on May 5 as ChatGPT’s new default model, replacing GPT‑5.3 Instant for all users. (openai.com) - The headline number is 52.5% fewer hallucinated claims on high‑stakes prompts, plus 37.3% fewer inaccurate claims on conversations users had flagged. (openai.com) - It matters because ChatGPT’s default model is the one most people actually use — and paid tiers still get manual model selection. (openai.com)
ChatGPT’s default model just changed, and the point is simple: make the fast version less likely to make things up when the topic actu(openai.com)5 Instant as the new default in ChatGPT, replacing GPT‑5.3 Instant. The pitch is not “more genius.” It’s “more dependable” — especially on medicine, law, (openai.com)slower one. (openai.com) ### What actually changed? The everyday ChatGPT model — the on(openai.com) now GPT‑5.5 Instant. OpenAI says the update is available to everyone, while paid users still get the model picker and can manually choose other options like GPT‑5.5 Instant or GPT‑5.5 Thinking. (openai.com) ### Why “Instant” matters? “Instant” is the low-latency lane. It’s meant to answer quickly and handle the huge volume of normal ChatGPT use, so OpenAI can’t just swap in (openai.com) feel of the product. Basically, this launch is trying to keep the speed people expect while sanding down one of the biggest trust problems in chatbots. (openai.com) ### How much better is it supposed to be? The big number is 52.5% fewer hallucinated claims than GPT‑5.3(openai.com)nAI also says GPT‑5.5 Instant made 37.3% fewer inaccurate claims on especially hard conversations that users had previously flagged for factual errors. Those are internal tests, not a universal guarantee, but they tell you what the company optimized for. (openai.com) ### Why focus on “high-stakes” prompts? Because that’s where the pro(openai.com)ndation, annoying. If it confidently invents a legal rule, a drug interaction, or a financial fact, that can send people in the wrong direction fast. OpenAI is basically admitting that the default model now has to be judged less like a toy and more like infrastructure people lean on for real decisions. (openai.com) ### What’s the new memory piece? OpenAI also added “mem(openai.com)inds of personal context — like saved memories, past chats, or connected files — helped shape an answer, and they can control those sources. That is less flashy than the hallucination stat, but it matters because personalization only feels useful if people can see where it’s coming from. (openai.com) ### Does everyone get the same control? No — and that’s an important catch. Everyon(openai.com)is still gated by paid tiers. OpenAI’s help docs say Plus, Pro, and Business users have the model picker, while free users mostly experience whatever OpenAI sets as default and whatever automatic switching behavior the product chooses. (help.openai.com) ### Is this a bigger strategy shift? Yes. GPT‑5.5 launched in late April fo(openai.com) that family deeper into the main ChatGPT experience. Turns out the real story is not just a new model release — it’s OpenAI tuning the default consumer experience around reliability, personalization, and controlled transparency. (openai.com) ### Bottom line This is OpenAI fixing the part of AI that burns trust fastest: fast a(help.openai.com)sting, the default ChatGPT experience gets a little less magical — and a lot more usable. (openai.com)
