OpenAI has released GPT-5.3 Instant, an update to the most widely used ChatGPT model, focused on improving the quality and flow of everyday conversations. The new version reduces unnecessary refusals and hallucinations while delivering more direct, relevant answers, changes that affect how people use the system for work, research, and creative tasks.
The update centers on tone, relevance, and conversational flow, areas that shape whether responses feel helpful or frustrating in routine use.
SEE ALSO: OpenAI changes deal with US Department of War after backlash
OpenAI said feedback showed GPT-5.2 Instant sometimes declined questions it could have answered safely. In other cases, responses included cautious or moralizing preambles that slowed down the exchange.
GPT-5.3 Instant reduces those unnecessary refusals and cuts back on defensive language before answering. When a question can be addressed appropriately, the model now responds directly and stays focused on the request. This should lead to fewer dead ends and answers that feel overly restrictive.
GPT-5.3 Instant reasoning
The model also changes how it handles information from the web. GPT-5.3 Instant balances online results with its internal knowledge and reasoning instead of leaning heavily on search output alone.
This approach affects how recent events and developing topics are explained. Rather than presenting long lists of links or loosely connected summaries, the model integrates key details into a more cohesive answer.
Earlier versions sometimes over-indexed on web content, which could produce fragmented responses. GPT-5.3 Instant places the most relevant information near the beginning of its replies while maintaining speed and clarity. Tone was another area addressed in the update.
Some users described GPT-5.2 Instant as being overbearing or prone to making assumptions about intent or emotions. The newer model adopts a more restrained and natural conversational style, removing exaggerated phrasing and unnecessary proclamations.
OpenAI is also working to keep the assistant’s personality consistent across updates, so improvements in capability won’t disrupt familiarity. Users can adjust response tone, including warmth and enthusiasm, within settings.
Beyond conversational style, GPT-5.3 Instant shows gains in factual accuracy. OpenAI evaluated the model using two internal benchmarks focused on reliability.
One assessment covered higher-stakes domains including medicine, law, and finance. The second measured hallucination rates in de-identified ChatGPT conversations that users flagged for factual errors, which tend to be especially prone to mistakes.
In the higher-stakes evaluation, GPT-5.3 Instant reduced hallucination rates by 26.8 percent when using the web and 19.7 percent when relying only on internal knowledge compared to prior models. In the user-feedback evaluation, hallucinations decreased by 22.5 percent with web access and 9.6 percent without it.
OpenAI said GPT-5.3 Instant also performs better as a writing partner. It produces clearer and more immersive prose when drafting fiction, refining passages, or exploring ideas. The model moves between practical tasks and expressive writing without losing coherence or focus.
Although the update improves everyday usability, some challenges remain. In certain non-English languages, including Japanese and Korean, responses can still sound literal or stilted. Improving tone and naturalness across languages remains an ongoing area of work.
GPT-5.3 Instant is now available to all ChatGPT users and to developers through the API under the name gpt-5.3-chat-latest. Updates to Thinking and Pro models are also planned.
What do you think about GPT-5.3 Instant? Let us know in the comments.
Image credit: rarrarorro / depositphotos
