OpenAI announced on May 5, 2026 that ChatGPT's default model is being switched to "GPT-5.5 Instant." Compared with the previous GPT-5.3 Instant, the new model is said to hallucinate substantially less in high-stakes domains such as medicine and law, while answers are shorter and the conversational tone feels more natural. This article walks through what changed and what it means for both ChatGPT users and developers.

A new default model and a new API alias

GPT-5.5 Instant rolls out today to every ChatGPT user, replacing GPT-5.3 Instant as the default. OpenAI says that because Instant is "the daily driver for hundreds of millions of people," even small changes make a big difference in everyday experience, and this update focuses on three axes: accuracy, conciseness, and personalization[1].

For developers, the new model is available via the API immediately under the model ID "chat-latest." This is a dynamic alias that always points to the latest Instant-class model, so applications get the upgrade without any explicit model-name change. GPT-5.3 Instant remains selectable from the model settings for paid plan subscribers for three more months before being retired[1][2].

Higher accuracy in high-stakes domains and shorter answers

On the accuracy front, the headline numbers come from evaluations in domains where factual errors carry the highest cost: medicine, law, and finance. According to OpenAI's internal evaluations, GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on these high-stakes prompts. On a separate, particularly hard set of conversations that users had previously flagged for factual errors, inaccurate claims were reduced by 37.3%[1][3].

The tone and length of replies have also been reworked. In OpenAI's own published comparison, the new model used 30.2% fewer words and 29.2% fewer lines for the same prompt. Gratuitous emojis and boilerplate openings are reduced, and the model is described as asking fewer "just to check" follow-up questions and producing less heavy formatting[1][3]. OpenAI further says general capabilities are up across image-based reasoning, STEM questions, and the model's judgment of when to invoke web search[1].

Personalization, memory sources, and rollout schedule

The other headline change is that GPT-5.5 Instant can pull context from past chats, uploaded files, and a connected Gmail account when it judges that personalization will help. The intent is to surface this context only when it improves the answer, so users do not have to keep restating prior assumptions[1][4].

Tied to this, every ChatGPT model is gaining a "memory sources" feature. When a response is personalized, ChatGPT shows which saved memories or past chats were referenced, and users can delete or correct stale entries on the spot. Memory sources are not displayed in shared chats, and temporary chats let users opt out of memory reads and writes entirely. OpenAI notes that memory sources will not necessarily surface every factor that shaped an answer, and that this view will keep improving over time[1].

Rollout is staged. Enhanced personalization starts with Plus and Pro users on the web, expands to mobile soon, and is planned to reach Free, Go, Business, and Enterprise users in the coming weeks. Available personalization sources may vary by region. Memory sources begin rolling out across all consumer plans on the web, with mobile to follow shortly[1].

Summary

OpenAI has positioned GPT-5.5 Instant as the new default ChatGPT model, claiming a 52.5% reduction in hallucinated claims on high-stakes prompts and 30.2% shorter responses on internal evaluations. It is available through the API as chat-latest from day one, while GPT-5.3 Instant remains accessible to paid users for three more months. Because every figure here comes from OpenAI's own evaluations, the practical answer for most teams is the same as ever: try it on your real workloads before drawing conclusions.

出典:[1] OpenAI "GPT-5.5 Instant: smarter, clearer, and more personalized" (https://openai.com/index/gpt-5-5-instant/)

出典:[2] TechCrunch "OpenAI releases GPT-5.5 Instant, a new default model for ChatGPT" (https://techcrunch.com/2026/05/05/openai-releases-gpt-5-5-instant-a-new-default-model-for-chatgpt/)

出典:[3] 9to5Mac "GPT-5.5 Instant makes ChatGPT more accurate while nixing 'gratuitous emojis'" (https://9to5mac.com/2026/05/05/gpt-5-5-instant-makes-chatgpt-more-accurate-while-nixing-gratuitous-emojis/)

出典:[4] Neowin "OpenAI upgrades ChatGPT's default Instant model with better accuracy and personalization" (https://www.neowin.net/news/openai-upgrades-chatgpts-default-instant-model-with-better-accuracy-and-personalization/)