OpenAI Launches GPT-5.5 Instant as New Default ChatGPT Model

OpenAI has officially released GPT-5.5 Instant, a new lightweight AI model that now serves as the default experience inside ChatGPT. The launch comes shortly after the company introduced GPT-5.5 and after Microsoft added the model family to Microsoft Foundry.

The new GPT-5.5 Instant replaces GPT-5.3 Instant as the standard model for all ChatGPT users. OpenAI says the updated model delivers more accurate answers, a more natural conversational style, and improved reasoning performance, especially in STEM-related tasks and image understanding.

GPT-5.5 Instant focuses on speed and accuracy

According to OpenAI, GPT-5.5 Instant reduces hallucinations by 52.5% compared to GPT-5.3 Instant. The company also shared several benchmark results showing gains across reasoning and multimodal tasks.

The model reportedly scored 85.6% on GPQA, 81.2% on AIME 2026, and 76% on MMMU-Pro. OpenAI says these improvements should make ChatGPT responses more reliable while keeping the fast response times users expect from the Instant model lineup.

OpenAI also refined the conversational tone of the model. Responses should now feel shorter, clearer, and more natural during regular chats.

Personalization and memory features are expanding

Another major change involves ChatGPT personalization. OpenAI introduced a new “memory sources” section that shows users exactly what information ChatGPT uses for context.

This includes memories, previous chats, uploaded files, and integrations such as Gmail. Users can also edit or remove stored memories directly from the interface.

The company says the personalization upgrades currently roll out first to Plus and Pro users on the web. Support for mobile apps and additional plans, including Free, Go, Business, and Enterprise, will expand later.

GPT-5.3 Instant will remain temporarily available

While GPT-5.5 Instant becomes the default model for ChatGPT, OpenAI confirmed GPT-5.3 Instant will remain available for paid subscribers for the next three months.

Developers can also access the new model through the API under the “chat-latest” identifier.

OpenAI additionally announced GPT-5.5-Cyber, a separate model designed to assist cybersecurity professionals with research and defensive workflows.

Via Neowin