ChatGPT now defaults to GPT‑5.5 Instant to reduce hallucinations

ChatGPT’s default model just changed, and the point is pretty simple — make the thing lie less. On May 5, OpenAI started rolling out GPT‑5.5 Instant as the default model inside ChatGPT, replacing GPT‑5.3 Instant for everyday use. The pitch is not “here’s a flashy new supermodel.” It’s “the model most people touch every day should be tighter, calmer, and less likely to make things up.” That matters more than it sounds, because defaults are the product. ### What actually changed? The change is at the ChatGPT product layer. GPT‑5.5 Instant is now the default model for all ChatGPT users, and OpenAI says it replaces GPT‑5.3 Instant across everyday chats. Paid users still get a model picker and can manually choose other options, but most people will just feel the new default without doing anything. (openai.com) ### Why does the default matter so much? Because most users never touch the picker. They use whatever shows up when they open ChatGPT. So a default-model swap is not a niche enthusiast update — it changes the behavior of the product at mass scale. OpenAI framed Instant as the “daily driver” for hundreds of millions of people, which is another way of(openai.com)el is the front door. (openai.com) ### What is GPT‑5.5 Instant supposed to be better at? Basically three things — factuality, concision, and personalization. OpenAI says answers are more accurate, more concise, better formatted, and more natural in tone. The company also says the model handles image understanding, STEM questions, and web-search decisions better than the previous defa(openai.com)p. It is the fast one that is supposed to be dependable. (openai.com) ### How much less does it hallucinate? The strongest claim is the 52.5% figure. In OpenAI’s internal evaluations, GPT‑5.5 Instant produced 52.5% fewer hallucinated claims than GPT‑5.3 Instant on high-stakes prompts in medicine, law, and finance. OpenAI also says it cut inaccurate claims by 37.3% in especially difficult conversations users had previo(openai.com)rs, so they are best read as directional, but they tell you what OpenAI is optimizing for. (openai.com) ### What’s the “memory sources” part? OpenAI also added a way to show which saved memories or chat-history details influenced a response. That feature is called memory sources. The idea is transparency — if ChatGPT sounds unusually tailored, you can see whether it pulled that from stored context rather than from the current prompt alone. That does n(openai.com)rstand why the model answered the way it did. (openai.com) ### Is this the same as GPT‑5.5 Thinking? No — and that distinction matters. Instant is the quick everyday model. Thinking is the slower, deeper-reasoning option. OpenAI says paid users can choose directly, and ChatGPT can also automatically switch from Instant to Thinking for more complex tasks if that setting is enabled. So “GPT‑5.5” is a family name here, not one single behavior. (help.openai.com) ### Why does this matter beyond one model update? Because it shows where the product is heading. For a while, AI launches were mostly about peak capability — bigger benchmarks, harder tasks, longer chains of reasoning. But the thing ordinary users notice first is whether the assistant sounds confident and wrong. (help.openai.com), cleaner formatting, clearer personalization, and fewer made-up claims are all part of making failure feel more predictable. (openai.com) ### Bottom line? This is a trust update disguised as a model update. GPT‑5.5 Instant matters because it changes the version of ChatGPT most people actually use — and the main promise is not brilliance, but fewer unforced errors. (openai.com)