ChatGPT Adopts GPT-5.5 Instant: Fewer Errors, More Conciseness

⚡

Key Takeaways

1On May 5, 2026, OpenAI introduced GPT-5.5 Instant as the default model for ChatGPT, replacing GPT-5.3 Instant.

2GPT-5.5 Instant reduces factual errors on complex queries by 52.5%, according to OpenAI.

3The new Memory Sources provide users with insight into the data influencing the chatbot's responses.

💡Why it matters — This update enhances the accuracy and transparency of ChatGPT, strengthening its reliability for critical use cases.

A New Model for ChatGPT

On Tuesday, May 5, 2026, OpenAI announced the integration of GPT-5.5 Instant as the new default model for ChatGPT. This model succeeds GPT-5.3 Instant, which was deployed in early March. GPT-5.3 Instant will remain available for paying subscribers for three months before being retired. OpenAI presents GPT-5.5 Instant as being more reliable for everyday tasks and in areas where accuracy is crucial.

Significant Reduction in Errors

According to OpenAI's internal assessments, to be taken with caution, GPT-5.5 Instant reduces factual errors by 52.5% compared to its predecessor on complex queries, particularly in medical, legal, and financial domains. Additionally, conversations reported for inaccuracies see their error rate decrease by 37.3%. OpenAI illustrates this improvement with an example: when faced with an algebraic problem containing an error, GPT-5.5 Instant detects the incorrect solution, identifies the calculation mistake, and recalculates the correct result. In contrast, GPT-5.3 Instant also spotted the error but incorrectly concluded that no solution existed.

More Concise and Relevant Responses

The new model offers shorter responses, with fewer emojis, while maintaining the warmth and personality of ChatGPT. It also reduces unnecessary follow-up questions and utilizes the context of past conversations, shared files, and even Gmail if that messaging service is connected, more effectively. OpenAI emphasizes that GPT-5.5 Instant "gets to the point more without losing substance."

Introduction of Memory Sources

One of the new features of GPT-5.5 Instant is the integration of Memory Sources. This functionality allows users to see the context used to personalize a response. A new sidebar displays the memories or past conversations that influenced the response, similar to sources in web research. Users can mark these snippets as relevant, irrelevant, or delete them. OpenAI clarifies that "Memory Sources are intended to make personalization more transparent, but they do not necessarily reflect all the factors that influenced a response."

Gradual Rollout

Since May 5, GPT-5.5 Instant has been accessible to all ChatGPT users. Developers can also use it via the chat-latest API. Advanced personalization will first be available to Plus and Pro subscribers on the web, before expanding to iOS, Android, and other plans in the coming weeks.

Having learned from the negative reactions during the withdrawal of GPT-4o in 2025, OpenAI will maintain GPT-5.3 Instant for paying subscribers for three months.