OpenAI Cuts Hallucinations by 52% With GPT-5.5 Instant Rollout

OpenAI has replaced ChatGPT's default model with GPT-5.5 Instant, rolling out the change to all users simultaneously while also making the model available in the API as "chat-latest." The update marks a significant shift in how the company is positioning its flagship consumer product, prioritizing accuracy and personalization over benchmark scores.

Hallucination Rates Drop on High-Stakes Prompts

The headline claim from OpenAI is a 52.5% reduction in hallucinated claims compared to GPT-5.3 Instant, measured on high-stakes prompts in medicine, law, and finance. On conversations that users had previously flagged for factual errors, the new model reduced inaccurate claims by 37.3%.

Control vs. Scale: How Anthropic and OpenAI Diverged on Cyber AI Risk

Anthropic limits cyber AI access via Project Glasswing. OpenAI scales broadly. Which risk philosophy fits your organization best?

Mission MediaMission Media

OpenAI said the improvements reflect the industry's continued focus on trust and reliability as AI developers try to make consumer tools more useful for everyday work and personal tasks. Paid users retain access to GPT-5.3 Instant for three months through model settings before it is withdrawn.

The company described GPT-5.5 Instant as the version used every day by hundreds of millions of people. At that scale, even incremental accuracy gains have a broad impact across study, work, and routine information-seeking.

Memory Sources and Personalization Controls

Alongside the model change, OpenAI introduced what it calls memory sources across all ChatGPT consumer plans. The feature shows users what context shaped a personalized response, including saved memories, past chats, uploaded files, and connected Gmail accounts.

Users can delete chats they no longer want referenced, remove or edit saved memories in settings, or switch to temporary chats that do not use or update memory. OpenAI noted that memory sources will not be visible when a chat is shared with other people.

GPT-5.5 Instant is better at drawing on context from earlier conversations, uploaded files, and Gmail when those options are enabled, OpenAI said. The improvement is intended to reduce the need for users to repeat background information and to support more tailored suggestions and follow-up tasks.

OpenAI acknowledged the feature is not yet comprehensive. In some cases it will display only the most relevant past chats rather than every conversation the system searched. "We'll continue improving this view to make it more comprehensive over time," the company said.

Behavioral Changes in Day-to-Day Use

Beyond accuracy, the new model is designed to produce tighter responses with less unnecessary formatting while maintaining a conversational tone. It asks fewer unnecessary follow-up questions and avoids excessive emoji use, which OpenAI framed as part of a broader push among AI providers to improve how systems behave in daily use rather than focusing only on headline model launches.

OpenAI's GPT-5.5 Signals a New Era for Marketing Teams

OpenAI's GPT-5.5 sharpens coding and research gains, reshaping how marketing teams compete on AI-driven workflows and strategy execution.

Mission MediaMission Media

The update has practical relevance for marketing and communications teams across Asia. The 52.5% reduction in hallucinated claims on medicine, law, and finance prompts is directly applicable to agencies drafting client materials in regulated sectors. The memory sources feature offers a new level of auditability, allowing professionals to see and correct the context driving personalized outputs before that content reaches clients or publication.

Staged Rollout

Enhanced personalization features, including improved use of past chats, files, and Gmail, are being introduced first to Plus and Pro subscribers on the web. Mobile support follows, with a broader expansion to Free, Go, Business, and Enterprise plans after that. Memory sources are rolling out across all ChatGPT consumer plans on the web, with mobile support to follow. Regional availability of specific features may vary.