OpenAI Makes GPT-5.5 Instant the Default ChatGPT Model

The Story

OpenAI is rolling out GPT-5.5 Instant as the new default ChatGPT model for all users and the API, where it is available as chat-latest. The company says the update improves accuracy, everyday task performance, personalization and response style while maintaining the low latency of GPT-5.3 Instant. Enhanced personalization is initially available to Plus and Pro users on the web, with mobile access and wider rollout to Free, Go, Business and Enterprise coming later.

The Facts

• GPT-5.5 Instant is replacing GPT-5.3 Instant as ChatGPT’s default model.

  The rollout starts today for all ChatGPT users and the API, where the model is available as chat-latest.

• The update focuses on accuracy, personalization and clearer responses.

  OpenAI says GPT-5.5 Instant gives stronger and tighter answers across subject areas, uses more relevant user context and avoids clutter such as gratuitous emoji and excessive formatting.

• The model keeps the low latency of its predecessor.

  OpenAI says GPT-5.5 Instant reduces hallucinations in sensitive areas while maintaining the speed profile of GPT-5.3 Instant.

• The model improves several everyday capabilities.

  OpenAI says GPT-5.5 Instant is better at analyzing photo and image uploads, answering STEM-related questions and deciding when web search would make an answer more useful.

• Enhanced personalization starts with Plus and Pro users on the web.

  The feature uses past chats, files and connected Gmail, where available, to make answers more personally relevant. It is coming soon to mobile and later to Free, Go, Business and Enterprise in the coming weeks.

• Paid users can keep GPT-5.3 Instant temporarily.

  GPT-5.3 Instant will remain available for paid users for three months through model configuration settings before being retired.

• OpenAI reports fewer hallucinated claims.

  According to OpenAI, GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in areas such as medicine, law and finance.

• The company also reports fewer inaccurate claims in difficult conversations.

  OpenAI says the model reduced inaccurate claims by 37.3% in especially challenging user conversations that had been flagged for factual errors.

• OpenAI reports gains across visual reasoning, science, math and document parsing.

  In OpenAI’s evaluations, GPT-5.5 Instant improved from 75.0% to 81.6% on CharXiv-reasoning, 69.2% to 76.0% on MMMU-Pro, 78.5% to 85.6% on GPQA and 65.4% to 81.2% on AIME 2025. On OmniDocBench, its average error rate fell from 14.6% to 12.5%.

• The model is designed to match the scale of the task.

  For casual advice prompts, OpenAI says GPT-5.5 Instant gives more concise answers, asks fewer unnecessary follow-up questions and avoids overbuilt formatting.

• ChatGPT will draw on more user context when personalization helps.

  The model can use information from past chats, uploaded files and connected Gmail, and can use its search tool to refer back to relevant context.

• OpenAI is adding memory-source controls.

  Memory sources show users some of the context used to personalize an answer, such as saved memories or past chats. OpenAI says these sources are rolling out across all ChatGPT consumer plans on the web and soon on mobile.

• Users can manage or avoid memory use.

  OpenAI says users can delete or correct outdated memory, delete chats they no longer want cited, and use temporary chats that do not use or update memory. Memory sources are not shown to others when a chat is shared.

• Past model retirements have triggered user backlash.

  GPT-4o was deprecated in February 2026 after users objected to its removal, with some describing the model’s personality as emotionally important to them.

How to Access

GPT-5.5 Instant is rolling out to all ChatGPT users and to the API as chat-latest. Enhanced personalization is initially limited to Plus and Pro users on the web, with mobile access coming soon and expansion to Free, Go, Business and Enterprise planned in the coming weeks. Paid users can continue using GPT-5.3 Instant for three months through model configuration settings before it is retired. Availability of specific personalization sources may vary by region.

Benchmarks

OpenAI reports that GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts involving medicine, law and finance. The company also says it reduced inaccurate claims by 37.3% in difficult user conversations previously flagged for factual errors.

OpenAI also reports benchmark gains across scientific chart reasoning, expert multimodal reasoning, document parsing, PhD-level science and competition math. These results are reported by OpenAI, not cited in the source as independent third-party testing.

Risks / Limitations

The update gives ChatGPT access to more personal context when users allow it, including past chats, uploaded files and connected Gmail data. That may reduce repetition and make answers more useful, but it also increases the amount of personal or work information involved in chatbot interactions.

Memory sources are designed to make personalization easier to understand, but OpenAI says they may not show every factor that shaped an answer. For example, they may show some of the most relevant past chats rather than every past chat the model searched or referenced.

Lower hallucination rates also create a trust problem: users may become more confident in ChatGPT’s answers even though the model can still be wrong.

What to Watch Next

The main question is whether OpenAI’s memory-source controls are enough to satisfy users who want more personalized answers without feeling that ChatGPT is using too much of their private context.

Why This Matters

ChatGPT’s default model shapes the everyday AI experience for hundreds of millions of people. GPT-5.5 Instant could make ChatGPT more useful by reducing errors, preserving speed, improving concise answers and remembering more relevant context, but the same changes also sharpen the tradeoff between convenience, trust, privacy and user attachment to specific models.

This article was drafted with the assistance of generative AI. All facts and details were reviewed and confirmed by an editor prior to publication.