OpenAI has released an update to one of the most widely used models inside ChatGPT — GPT-5.3 Instant.
According to the company, the update aims to improve how ChatGPT answers questions, particularly by providing more accurate responses, better web-based results, and fewer unnecessary refusals or disclaimers that can interrupt conversations.
Fewer refusals, more direct answers and more
According to OpenAI, earlier versions of the model could sometimes refuse questions it could safely answer or respond with long explanations about safety rules before addressing the question.
GPT-5.3 Instant is designed to reduce these issues. The model now gives answers more directly when it is appropriate to do so, avoiding overly cautious language that can slow down conversations. This change is based on feedback from users who said earlier responses sometimes felt overly defensive or moralising.
Additionally, the update also improves how the model uses information from the internet. Instead of simply listing links or summarising web results, GPT-5.3 Instant is designed to combine online information with its own reasoning.
OpenAI says the model now does a better job of understanding the real intent behind a question and highlighting the most important information first.
More natural conversational tone
Another focus of the update is tone. Previous versions could sometimes sound overly formal or make assumptions about a user’s emotions. GPT-5.3 Instant is designed to respond in a more natural and consistent style, avoiding phrases or instructions that feel unnecessary or awkward during normal conversations.
Users can still adjust the tone of responses, including warmth and enthusiasm, through settings. Additionally, the company also reported improvements in factual reliability.
In internal evaluations, GPT-5.3 Instant showed lower hallucination rates, meaning the model produced fewer incorrect facts compared with earlier versions.
The company said hallucinations decreased by 26.8% when the model used web information and 19.7% when relying only on internal knowledge in tests covering fields such as medicine, law, and finance.
The new model is also designed to be a better writing assistant. OpenAI said GPT-5.3 Instant produces more expressive and detailed writing, allowing it to move more easily between practical tasks and creative writing. This includes generating fiction, editing text, and helping users develop ideas while keeping responses clear and structured.
Availability
GPT-5.3 Instant is available starting today for all ChatGPT users. Developers can also access it through OpenAI’s API under the model name “gpt-5.3-chat-latest.”
The previous model, GPT-5.2 Instant, will remain available for paid users in the legacy models section for three months and is scheduled to be retired on June 3, 2026.
