OpenAI makes GPT-5.5 Instant the default ChatGPT model, citing lower hallucination rates

OpenAI on May 5, 2026 released GPT-5.5 Instant and made it the default model in ChatGPT for all users, replacing GPT-5.3 Instant. The company positioned the model as a low-latency workhorse with materially improved accuracy: OpenAI said internal evaluations showed GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts spanning medicine, law and finance, while preserving the speed of its predecessor. On capability benchmarks, TechCrunch reported the model scored 81.2 on AIME 2025 math (versus 65.4 for the prior model) and 76 on the MMMU-Pro multimodal suite (versus 69.2). The release also extended ChatGPT memory: GPT-5.5 Instant can use its search tool to refer back to past conversations, uploaded files and connected Gmail to personalize answers, a feature rolling out first to Plus and Pro users on web.

Memory sources became visible across models with options to delete or correct them, and OpenAI described 'dreaming' functionality allowing stored memories to update automatically over time. For developers, the model is exposed via API as 'chat-latest,' with GPT-5.3 remaining selectable for paid users for three months before retirement. Paid ChatGPT users likewise retained access to GPT-5.3 Instant through model-configuration settings for the same window. The change is significant because it silently swaps the model the largest share of ChatGPT's user base interacts with daily.

Hallucination-reduction figures are OpenAI-reported and were not independently audited at filing. Reporting also noted unconfirmed rumors of a forthcoming GPT-5.6; as of early June 2026 OpenAI had not announced such a model.