Main points
- OpenAI updated ChatGPT to the GPT-5.5 Instant model, which increased response speed and reduced errors.
- GPT-5.5 Instant supports up to 128,000 context tokens, improves response personalization, and memory transparency by showing the sources of stored information.
OpenAI made ChatGPT more attentive to your chats and files / Depositphotos
OpenAI has quietly upgraded ChatGPT to the new GPT-5.5 Instant core model. The update has made the chatbot respond faster and make fewer mistakes. The platform also now has new capabilities for working with context and memory, which will be immediately noticeable in its work.
The GPT-5.5 Instant model replaces the previous one, GPT-5.3 Instant. Now, version 5.5 has become the base for ChatGPT. As TechCrunch reports, the company claims that the new version better copes with so-called “hallucinations” – cases when artificial intelligence generates convincing but false information.
OpenAI is focusing on sensitive topics such as medicine, finance, and law, while maintaining the low latency of the previous Instant model.
Why did OpenAI change the core ChatGPT model again?
GPT-5.5 was first introduced back in April. At the time, the company announced improvements in programming, working with knowledge bases, and logical tasks. Now, these changes have also made their way to the main version of ChatGPT.
According to OpenAI, the new model significantly improved results in a number of tests.
- In the AIME 2025 math benchmark, GPT-5.5 Instant scored 81.2 points compared to 65.4 for GPT-5.3 Instant;
- In the multimodal MMMU-Pro test, the result also increased – from 69.2 to 76 points.
One of the main innovations is working with context . GPT-5.5 Instant can use the search tool to access previous chats, files, and Gmail integrations to generate more personalized responses.
In the top tier, GPT-5.5 Instant supports up to 128,000 context window tokens , and the GPT-5.5 Thinking version supports up to 400,000 . This allows the model to work with large documents, long correspondence, or large files without losing context.
The feature will initially be available to Plus and Pro users on the web version of the service, and will later be available on mobile apps. OpenAI also said that the feature will be available to users of the free plan, as well as Go, Business, and Enterprise subscriptions in the coming weeks.
Another notable change concerns ChatGPT's memory transparency . The service will now show the sources from which the model takes saved context for results. Users will be able to delete outdated data or correct it if the system used incorrect information.
At the same time, OpenAI emphasizes that people to whom a user sends a chat link will not see these memory sources.
GPT-5.5 will be available to developers via an API called “chat-latest” . The company will keep the old GPT-5.3 Instant for paid users for about three more months as an alternative.
What immediate changes will ChatGPT users notice?
OpenAI claims that GPT-5.5 Instant is not just “smarter” but also significantly more practical in everyday use. According to the company's internal tests, the new model generates 52.5% fewer “hallucinations” on complex topics. It is also 37.3% less likely to give inaccurate information in dialogues that users have previously marked as problematic.
ChatGPT will now provide shorter and “cleaner” answers without excessive emojis and overly emotional wording. The model will better adapt to the complexity of the query – simple questions are processed quickly, while more complex ones can automatically receive deeper analysis, writes The Verge.
ChatGPT as a personal assistant
It is noticeable that OpenAI emphasizes personalization of ChatGPT. The company essentially transforms the chatbot from a disposable tool into a permanent digital assistant that remembers the user's habits, previous conversation topics, and related content.
That is why OpenAI has separately launched the Memory Sources feature, which should show where the model got the information for a personalized response.
OpenAI has recently come under fire after it shut down its GPT-4o model. After it was shut down, some users reacted negatively to OpenAI's decision, as they had already gotten used to the model's “personality.”
Some users even called GPT-4o their best friend. Fans of the model even started collecting signatures on the Change.org platform to bring GPT-4o back. However, despite criticism and petitions, OpenAI finally stopped supporting GPT-4o in February 2026.