OpenAI releases GPT-5.3 Instant with fewer refusals and improved web answers
New ChatGPT model improves conversational flow, reduces hallucinations, and delivers more useful results when combining web search with internal knowledge.
Share
Add us on Google by Estefano Gomez Mar. 3, 2026OpenAI today released GPT-5.3 Instant to all users of ChatGPT, introducing a new version of its most widely used conversational model designed to deliver more accurate responses and smoother everyday interactions.
The model is now available across ChatGPT on web, mobile, and desktop, replacing the previous GPT-5.2 Instant model for most users.
According to OpenAI, the update focuses on improving conversational flow, response relevance, and overall usability. The company said the model delivers more accurate answers, produces better synthesized responses when using web information, and reduces overly cautious language that can interrupt conversations.
One of the main changes is improved judgment around refusals. OpenAI said GPT-5.2 Instant sometimes declined questions it could safely answer or responded with defensive disclaimers. GPT-5.3 Instant reduces unnecessary refusals and minimizes overly cautious preambles so answers are delivered more directly.
The model also improves how ChatGPT integrates web search results. Rather than listing links or loosely connected information, GPT-5.3 Instant is designed to combine web data with its own knowledge to produce more relevant and contextualized responses.
The update also improves writing performance. OpenAI said GPT-5.3 Instant provides more natural and expressive writing while maintaining clarity across practical tasks and creative content.
GPT-5.3 Instant is available immediately to all ChatGPT users and to developers through the API under the model name gpt-5.3-chat-latest.
The previous GPT-5.2 Instant model will remain available to paid users for three months in the legacy model picker before being retired on June 3, 2026.
Disclosure: This article was edited by Estefano Gomez. For more information on how we create and review content, see our Editorial Policy.