OpenAI swaps in GPT-5.5 Instant

ChatGPT’s default model just changed, and this one is less about flashy new powers than about making the thing people already use every day feel more trustworthy. OpenAI is rolling GPT-5.5 Instant into ChatGPT as the new default, replacing GPT-5.3 Instant for general use. (openai.com) The pitch is simple — fewer made-up answers, cleaner responses, and better use of what you’ve told ChatGPT about yourself when personalization actually helps. OpenAI posted the update on May 5, 2026, and the change is now showing up in ChatGPT release notes and help docs. ### What actually changed? (help.openai.com) The big product change is that GPT-5.5 Instant is now the default model in ChatGPT for everyday use. OpenAI says it replaces GPT-5.3 Instant for all ChatGPT users, not just paid plans, while paid users still get a model picker that can manually choose GPT-5.5 Instant or GPT-5.5 Thinking. ### Why “Instant” matters “Instant” is the fast everyday mode — the one meant for normal chats, quick questions, image uploads, and routine work. That matters because this is the model hundreds of millions of people hit first, so even small gains land everywhere. OpenAI is framing GPT-5.5 Instant as a daily-driver upgrade, not a niche power-user model. (openai.com) ### Is it just faster, or actually better? OpenAI is leaning hard on factuality. In its own testing, 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.com) It also cut inaccurate claims by 37.3% on difficult conversations users had flagged for factual errors. Those are internal numbers, so take them as product claims, but they tell you what OpenAI thinks the headline improvement is. ### What about the answers themselves? The company also says outputs are more streamlined — less header clutter, cleaner formatting, and a more natural tone. (openai.com) That sounds cosmetic, but it is actually a meaningful product choice. A lot of frustration with chatbots comes from answers that are technically fine but padded, repetitive, or weirdly over-structured. OpenAI is clearly trying to make ChatGPT feel less like it is performing “assistant voice” and more like it is just answering. ### Where does deeper reasoning fit? This is the more interesting architecture shift. In ChatGPT, selecting Instant does not always mean one fixed model anymore. (openai.com) OpenAI’s help docs say Instant can automatically decide whether to use GPT-5.3 Instant or switch to GPT-5.5 Thinking for harder requests, applying deeper reasoning before answering. Basically, the fast lane can quietly hand off to the slower, more deliberate lane when the task needs it. ### Can users control that switch? Yes — at least on paid tiers. The model picker now lets Plus, Pro, and Business users turn automatic switching between Instant and Thinking on or off, access legacy models, and set thinking effort when they choose Thinking or Pro. (openai.com) OpenAI also says GPT-5.5 Thinking and Pro may start with a short preamble when they begin reasoning, which makes the “thinking” step more visible than before. ### Why bring memory into this? Because personalization cuts both ways. OpenAI says GPT-5.5 Instant has improved personalization controls, and its memory docs still emphasize that users can inspect, delete, clear, or disable saved memory entirely. (help.openai.com) So the upgrade is not just “the model remembers more” — it is also “you’re supposed to have clearer control over what sticks.” ### Bottom line This is a quality-of-life release, but those often matter more than splashy demos. If GPT-5.5 Instant really makes default ChatGPT noticeably less wrong and less verbose, that changes the product people actually touch every day — not just the benchmark story around it. (help.openai.com) (openai.com)