OpenAI has released GPT-5.3 Instant, a new version of the model that powers much of the day-to-day ChatGPT experience. The focus this time is something more practical than just reasoning benchmarks; it’s focused on improving reliability, response speed, and conversational quality.
OpenAI promises a 26 percent reduction in hallucinations
According to OpenAI, GPT-5.3 Instant reduces hallucinations by 26.8 percent when using the web and 19.7 percent when relying only on its internal knowledge, based on internal evaluations across areas like medicine, finance, and law.
Hallucinations are still one of the most persistent and annoying issues with generative AI systems. A model that can generate text quickly is useful, but one that confidently just makes up information is obvious and much harder to deploy in real workflows.
OpenAI says the goal with GPT-5.3 Instant is to make everyday interactions with ChatGPT feel smoother and more reliable.
Improving everyday chatter
The update also focuses on how ChatGPT behaves in conversation, not just performance.
“This update focuses on the parts of the ChatGPT experience people feel every day: tone, relevance, and conversational flow,” said OpenAI in its announcement. “These are nuanced problems that don’t always show up in benchmarks, but shape whether ChatGPT feels helpful or frustrating. GPT‑5.3 Instant directly reflects user feedback in these areas.”
Earlier versions of the model were also sometimes slammed for generating responses that felt overly cautious or padded with long disclaimers. GPT-5.3 Instant is designed to skip some of that.
“GPT‑5.3 Instant significantly reduces unnecessary refusals, while toning down overly defensive or moralizing preambles before answering the question,” states OpenAI. “When a useful answer is appropriate, the model should now provide one directly, staying focused on your question without unnecessary caveats. In practice, this means fewer dead ends and more directly helpful answers.”
The model also does a better job blending information it finds online with stuff it already knows.
GPT-5.3 Instant doesn’t just list links or summarize search results; it pulls all of that information into one single answer. The goal is simpler responses that get to the point faster.
Accuracy, my dear Watson
The update also says something about how OpenAI is talking about new models. The focus is less on benchmark scores and more on reliability and usability.
GPT-5.3 Instant is currently the most widely used model in ChatGPT, so even smaller changes are noticeable.
The model is now available in ChatGPT and via the API under the name “gpt-5.3-chat-latest.” GPT-5.2 Instant will still be available for a few months before being phased out.
For developers building AI features into software, this is quite a boon, as faster responses and fewer hallucinations often matter more than benchmark scores.
DeepKeep this week introduced an AI Agent Scanner designed to map the growing attack surface created as AI agents plug into business systems and workflows. As more organizations embed AI into everyday software, visibility and reliability around those systems are becoming just as important as the models themselves.