'GPT-5.5 Instant' is now available, becoming the new default model for ChatGPT.



On May 5, 2026, OpenAI announced ' GPT-5.5 Instant ' as the new default model for ChatGPT. GPT-5.5 Instant replaces GPT-5.3 Instant and features improved accuracy, conciseness, natural conversation, and user-specific responses.

GPT-5.5 Instant: smarter, clearer, and more personalized | OpenAI

https://openai.com/index/gpt-5-5-instant/

GPT-5.5 Instant System Card | OpenAI
https://openai.com/index/gpt-5-5-instant-system-card/

GPT-5.5 Instant is the latest version of the 'Instant' series of models used daily in ChatGPT. OpenAI explains that because Instant has become the basic model for hundreds of millions of users on a daily basis, even small improvements can have a big impact.

One of the biggest improvements is the increased factual accuracy. OpenAI's internal evaluation showed that GPT-5.5 Instant reduced false claims by 52.5% compared to GPT-5.3 Instant in prompts in high-risk areas where accuracy is particularly important, such as healthcare, law, and finance.



Furthermore, it is reported that in previous models, even in complex conversations that users reported as factual errors, inaccurate claims were reduced by 37.3%. System cards also show a significant improvement in factual accuracy with GPT-5.5 Instant compared to GPT-5.3 Instant, with particularly noticeable improvements in high-risk areas.

In terms of capabilities, performance has improved across a wide range of everyday tasks, including analyzing uploaded photos and images, answering questions in STEM fields, and making decisions about necessary web searches. The evaluation results show increases in CharXiv-reasoning from 75.0% to 81.6%, MMMU-Pro from 69.2% to 76.0%, GPQA from 78.5% to 85.6%, and AIME 2025 from 65.4% to 81.2%.



The style of responses has also been revised, allowing for shorter, more concise answers that get straight to the point while maintaining informational value. OpenAI explains that GPT-5.5 Instant reduces unnecessary length, excessive formatting, unnecessary follow-up questions, and inappropriate emojis, resulting in more practical and natural responses.

Furthermore, personalization features have been enhanced, allowing GPT-5.5 Instant to more effectively leverage the context of past chats, files, and connected Gmail accounts, making it easier to continue work in progress and provide suggestions tailored to the user's preferences.

In addition, 'memory sources' will be introduced across ChatGPT. This is a mechanism that allows users to review the context of their past chats and the memory used to personalize their responses, and to delete or modify outdated or unnecessary information.



In terms of security, GPT-5.5 Instant is the first Instant-based model to be rated 'High Capability' in the Cybersecurity and Bio-Chemical Readiness categories. OpenAI explains that based on this assessment, it has applied bio-chemical and cybersecurity safeguards.

The graph below shows how well the model can withstand aggressive prompts aimed at bypassing security measures, known as 'jailbreaks.' The vertical axis, 'worst-case defender success rate,' indicates the percentage of times the model avoided harmful instructions; a higher rate means the security measures are less likely to be breached. The horizontal axis, 'attacker budget,' represents how many attempts and ingenuity the attacker can employ; the further to the right, the more difficult the attack conditions become.



GPT-5.5 Instant will become the default model for all ChatGPT users, replacing GPT-5.3 Instant, and will be rolled out gradually starting May 5th. It will be available as 'chat-latest' in the API, and paid users will be able to use GPT-5.3 Instant for three months from the model settings, after which it will be discontinued.

OpenAI CEO Sam Altman commented on the release of GPT-5.5 Instant, saying, 'The new instant model of ChatGPT is really good! If you've been using only the Thinking model for a while, you should try it! The improvements in speed, intelligence, personality, and superior memory and personalization capabilities all come together to create an effect that is more than just the sum of its parts.'

Related Posts:

in AI,   Web Service, Posted by log1i_yk