OpenAI said on Tuesday it updated GPT-5.3 Instant, the most widely used model in its AI chatbot ChatGPT.

The update focuses on improving answer accuracy, enhancing web search integration and making conversations more natural.

On accuracy, the update notably reduces hallucinations. OpenAI said an internal evaluation in high-risk areas such as law and finance showed hallucinations fell 26.8 percent when using web search and 19.7 percent when using only internal knowledge. It also said an assessment based on anonymised user feedback showed hallucinations fell 22.5 percent with the web and 9.6 percent without it.

It also improved how web search is used. The updated model moves beyond simply summarising online search results and combines its own knowledge and reasoning to explain the context of the latest news and issues.

On conversation style, it improved unnecessary refusals to answer, excessive emotional expressions and making hasty assumptions about user intent. It also strengthened support for creative work such as drafting novels, revising sentences and exploring ideas.

GPT-5.3 Instant is available to all ChatGPT users. Developers can access it through the API under the model name "gpt-5.3-chat-latest". Updates to the Thinking model and the Pro model will also be applied sequentially.

The previous version, GPT-5.2 Instant, will remain available to paying users for three months under Legacy Models in the model selector, before the service ends on June 3.