OpenAI has rolled out GPT-5.3 Instant, an update designed to make ChatGPT feel more natural, more accurate, and less overbearing in everyday conversations. The company says users should notice fewer dramatic or preachy responses and a smoother flow when asking questions, especially during longer chats or web-based queries.
The update targets tone, relevance, and conversational rhythm, which many users flagged as weak points in GPT-5.2 Instant. That earlier version sometimes opened with long safety preambles or unnecessary refusals before answering a simple question. With GPT-5.3 Instant, OpenAI says the model now responds more directly and cuts down on defensive phrasing that interrupts the exchange.
OpenAI said in its release notes:
“GPT-5.3 Instant delivers more accurate answers, richer and better-contextualized results when searching the web, and reduces unnecessary dead ends, caveats, and overly declarative phrasing that can interrupt the flow of conversation.”
The company also acknowledged feedback that GPT-5.2 Instant could feel overly cautious. In response, GPT-5.3 Instant now shows “better judgment around refusals and fewer disclaimers,” while still following safety standards. As OpenAI put it:
“GPT-5.3 Instant significantly reduces unnecessary refusals, while toning down overly defensive or moralizing preambles before answering the question.”
Stronger Web Answers and Accuracy Gains
Beyond tone, GPT-5.3 Instant improves how it integrates web results into responses. Instead of listing links or loosely connected information, the model now blends online findings with its own reasoning, surfacing the most relevant details upfront.
OpenAI reports measurable accuracy gains. On internal evaluations in higher-stakes areas such as medicine, law, and finance, hallucination rates dropped by 26.8 percent when using the web and 19.7 percent without it. User-flagged factual errors also declined.
GPT-5.3 Instant is available now in ChatGPT and through the API as “gpt-5.3-chat-latest.” GPT-5.2 Instant will remain in the Legacy Models section for paid users until June 3, 2026, before retirement.
