OpenAI recently released an update to its most popular model, GPT-5.3 Instant, which supposedly provides more accurate answers, richer and better contextualized results during searches, and reduces the number of «dead ends», caveats, and overly declarative wording.
OpenAI recently released an update to its most popular model, GPT-5.3 Instant, which supposedly provides more accurate answers, richer and better contextualized results during searches, and reduces the number of «dead ends», caveats, and overly declarative wording.
«These nuances don’t always show up in benchmarks, but they determine whether ChatGPT seems useful or annoying. GPT-5.3 Instant directly reflects user feedback in these areas,» the company noted in its blog.
OpenAI also added that GPT-5.3 Instant «significantly reduces unnecessary rejections while mitigating overly defensive or moralizing introductory preambles before answering questions.»
«When a useful answer is appropriate, the model should now directly provide it, focusing on your question without unnecessary reservations. In practice, this means fewer „dead ends“ and more directly useful answers,» the company explained.
Where GPT‑5.2 Instant eventually answers the question, but struggles to explain its security limitations and starts with a long introduction, GPT‑5.3 Instant jumps straight to the answer.
Also, GPT‑5.3 Instant, as its developers write, improves the quality of responses when information comes from the network.
«It better recognizes the subtext of questions and brings to the fore the most important information, including explicit information, resulting in answers that are more relevant and immediately usable, without losing speed or changing tone,» OpenAI emphasized.
Thus, according to the company’s plan, GPT‑5.3 Instant answers appear more relevant to the user’s intent.
In addition, compared to previous models, GPT‑5.3 Instant provides more substantive answers, less «hallucinating» across a wide range of topics.
«We used two internal assessments to measure accuracy: one focused on more critical domains like medicine, law, and finance, and the other measured the frequency of hallucinations in depersonalized ChatGPT conversations,» OpenAI said about their research.
According to published data, in critical areas, GPT-5.3 Instant reduced the frequency of hallucinations by 26,8% when using the network and by 19,7% when relying only on its internal knowledge, compared to previous models. User feedback evaluation showed a reduction in hallucinations by 22,5% when using the network and by 9,6% without access to it.
GPT‑5.3 Instant is currently available to all ChatGPT users, as well as developers, in the API called ‘gpt-5.3-chat-latest.’ Updates for Thinking and Pro, the company says, will be coming soon.
Meanwhile, the number of ChatGPT uninstalls jumped 295% in a day — this is how users are reacting to OpenAI’s deal with the Pentagon.
We also recently reported that Wix announced a partnership with OpenAi to launch an app in ChatGPT. Now, users can create professional Wix Harmony-powered websites directly within the ChatGPT interface.
