ChatGPT’s New Default Model Is OpenAI’s GPT-5.5 Instant

GPT-5.5 Instant, a new foundation model released by OpenAI earlier today, will take the place of GPT-5.3 Instant as the default ChatGPT model. The company claims that the model lowers hallucinations in sensitive areas such as law, medicine, and finance, while retaining the same low latency as before.

With the release of the most recent GPT-5.5 model last month, OpenAI claimed to have made great progress in knowledge work and coding.

In the AIME 2025 math test, the new model scored 81.2, while the previous model scored 65.4. Additionally, with a score of 76 compared to 69.2 on the MMMU-Pro multimodal reasoning benchmark, it exceeded the previous version.

The context management was given special attention in the release. Also, in order to provide users and customers with more individualized responses, GPT-5.5 Instant may use its search feature to look up previous exchanges, files, and Gmail. The Plus and Pro customers will also have access to this functionality on the web, and mobile users will soon be able to use it. In the upcoming weeks, OpenAI intends to make this feature available to the enterprise, Go Business, and Free customers.

Also, with this upgrade, ChatGPT will also provide memory sources of varying kinds for all models so users can see where the replies are coming from, and users have the option to remove out-of-date sources or update them if a response is incorrect. The company also noted that sharing a chat will hide the memory sources from the recipient.

The GPT-5.5 model will be accessible to developers through the API as “chat-latest,” while 5.3 will only be available for three months as a premium user option.

The previous model withdrawal actions have been dismissed and rejected by the company. Users who were familiar with OpenAI’s GPT-4o model’s “personality” reacted strongly to the model’s withdrawal. Users also felt a connection to that specific model since GPT-4o regularly validated their selections. The model was commonly referred to as their “best friend” or “a mirror” by users who supported the petitions to prevent OpenAI from retiring it. In February 2026, GPT-4o was deprecated in spite of the controversy. 

Also, OpenAI has introduced several new features with user experience improvements. For ChatGPT Plus and Pro users on the web, there is improved personalisation that allows the model to refer to previous conversations, uploaded files, and connected Gmail accounts to deliver more relevant answers and feedback. A new “memory sources” feature, available across all models, lets users see exactly which past chats or files informed a personalized response. 

Additionally, GPT-5.5 Instant is designed in a way to be tighter and more direct, using roughly 30% fewer words while significantly reducing unwanted and unnecessary emojis and overformatting. Finally, this is the first instant-tier model classified as “High Capability” in cybersecurity and biological domains, which triggers automated safeguards typically reserved for more advanced models.

The rollout has started for all ChatGPT users on the web. Paid subscribers who prefer the older GPT-5.3 Instant model can still access it through their settings during a three-month transition period before it is finally retired.