OpenAI on Tuesday introduced GPT-5.5 Instant, an updated default model for ChatGPT that the company says improves factual accuracy while producing more concise replies. The company positioned the release as an incremental upgrade to its prior default, GPT-5.3 Instant, and published comparative performance figures from internal testing.
In OpenAI's internal evaluations, GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on so-called high-stakes prompts in medicine, law and finance. For conversations that users had previously flagged for factual errors, OpenAI reported a 37.3% reduction in inaccurate claims with GPT-5.5 Instant.
The company also cited benchmark gains. On the AIME 2025 competition math test, GPT-5.5 Instant's accuracy increased to 81.2% from 65.4%. The model scored 85.6% on GPQA, a PhD-level science evaluation, compared with 78.5% for the earlier version. Visual reasoning accuracy on CharXiv improved to 81.6% from 75.0%.
OpenAI highlighted changes to the model's style as well as its substance. The company said GPT-5.5 Instant uses 30.2% fewer words and 29.2% fewer lines in responses compared with GPT-5.3 Instant, characterizing the shift as a reduction in verbosity while retaining conversational tone and necessary content.
Alongside the model update, OpenAI announced a new "memory sources" capability that makes it transparent which past chats, uploaded files or connected data informed a personalized response. Users can view those memory sources and have controls to delete or correct the context the model referenced. OpenAI said that memory sources are not shared when users share chats with others.
The model can also search a user's past conversations and connected Gmail accounts to deliver tailored answers so users need not repeat information. OpenAI described the behavior as conditional - the system decides when added personalization materially improves the response.
GPT-5.5 Instant is rolling out to all ChatGPT users beginning Tuesday, and it will replace GPT-5.3 Instant as the default model. Paid users will retain access to GPT-5.3 Instant through configuration settings for three months before that version is retired. Developers can access the new model in the API under the identifier chat-latest.
Enhanced personalization that draws on past chats and connected Gmail accounts is initially available to Plus and Pro users on the web, with mobile access coming soon. OpenAI said it plans to widen the feature to Free, Go, Business and Enterprise plans in the coming weeks. Memory sources are being introduced across all consumer plans on the web and are expected on mobile shortly.
The company provided these performance metrics and rollout details as part of its product update. The numbers cited reflect OpenAI's internal testing and configuration choices tied to the ChatGPT product and API.