GPT-5.5 Instant

ChatGPT's default model, Instant, has been switched from GPT-5.3 Instant to GPT-5.5 Instant. It is rolling out to all users sequentially, and OpenAI explained that it aims for more accurate, concise, and personalized responses.

Accuracy has also improved. In OpenAI's internal evaluations, hallucinated claims on high-stakes prompts such as medical, legal, and financial topics decreased by 52.5% compared to GPT-5.3 Instant, and inaccurate claims in difficult conversations flagged by users as factual errors decreased by 37.3%. Performance also improved in photo/image analysis, STEM questions, judging when to use web search, and visual reasoning, math, and science evaluations.

The response style has been adjusted toward shorter, more focused answers. Verbose explanations, excessive formatting, unnecessary follow-up questions, and unnecessary emoji use have been reduced, while maintaining a natural tone suited to the situation.

Personalization has also been strengthened. By better utilizing past chats, files, and connected Gmail context, it reduces repeated explanations and more accurately reflects information users have already shared. Examples introduced include more sophisticated use of previous conversation history and preferences for local recommendations and tea recommendations.

In addition, a memory sources feature has been introduced across all ChatGPT models. Users can check which saved memories or past chats were used in a personalized response, and can delete or edit outdated or unwanted context.

The rollout scope is as follows.

• ChatGPT: GPT-5.5 Instant is being applied sequentially as the default model
• API: Available as chat-latest
• Paid users: Can keep using GPT-5.3 Instant for 3 more months
• Personalization feature: Rolling out sequentially starting with Plus/Pro on web, then expanding to mobile and Free/Go/Business/Enterprise