OpenAI just rolled out GPT-5.3 Instant, a major update to ChatGPT’s most-used model that aims to make conversations feel more natural and helpful. The big fix? The model now refuses fewer valid questions and drops those annoying, overly cautious preambles that made ChatGPT sound preachy or defensive. It also handles web search much better — instead of just dumping links, it blends online info with its own reasoning to give you more contextual answers. And yes, it’s finally ditching cringe phrases like “Stop. Take a breath.” The update also cuts down on hallucinations, especially in high-stakes areas like medicine and finance. OpenAI says these changes are directly based on user feedback, targeting the subtle but frustrating issues that make or break daily use. Why it matters: This update shows that improving AI isn’t just about benchmark scores — sometimes the biggest wins come from fixing the small, human-level annoyances that make a tool feel like a chore to use.