OpenAI launches GPT-5.5 Instant, a new virtualization model for ChatGPT

On Monday, OpenAI released a new base model called GPT-5.5 Instant, which will replace GPT-5.3 Instant as the default ChatGPT model. The company said the model reduces hallucinations in sensitive areas such as law, medicine and finance, while maintaining the low latency of its predecessor.

OpenAI released its latest GPT-5.5 model last month with the company claiming improvements in areas such as programming and cognitive work.

The new model also achieved a score of 81.2 on the AIME 2025 mathematics test, compared to 65.4 for the older model. It also outperformed its predecessor in the MMMU-Pro multimodal inference benchmark, receiving a score of 76.0 versus 69.2.

The release focused specifically on context management. GPT-5.5 Instant can use its search tool to reference previous conversations, files, and Gmail to provide you with more personalized answers. This feature will be available to Plus and Pro users on the web, with plans to roll out to mobile soon. OpenAI said it plans to expand access to this feature to Free, Go Business, and Enterprise users in the coming weeks.

With this update, ChatGPT will also look at memory resource usage across all forms to help you understand where answers are generated. Users can delete old sources or correct them if the answer is wrong. Most importantly, the company said that if you share a conversation with someone, they will not be able to see the memory sources.

For developers, the GPT-5.5 model will be available through the API as a “latest chat,” with 5.3 available as an option for paid users for only three months.

The company has faced pushback from previous recall steps. When OpenAI pulled its GPT-4o model, there was backlash from users who related to the “personality” of the model. GPT-4o repeatedly confirmed users’ choices, making them feel connected to that specific model. Users who signed petitions to prevent OpenAI from going out of business described the model as their “best friend” or “mirror.” Despite protests, GPT-4o was deprecated in February 2026.