OpenAI just made one of its BIGGEST move of 2026. And they barely made any noise about it. GPT-5.3 Instant just went live. It replaces the default ChatGPT model, the one most people talk to every single day. Here is what changed and why it matters more than any benchmark score. First, OpenAI admitted its own model was "cringe." Their word and not mine. GPT 5.2 had a habit of lecturing users, refusing harmless questions and opening with you're not broken before anyone asked. GPT-5.3 Instant kills that behavior. Fewer refusals and disclaimers. When the answer is safe to give, the model just gives it. OpenAI spent two years tightening safety guardrails. Sometimes too tight. And Users complained constantly, now the pendulum is swinging back. Second, hallucinations are down significantly. 26.8% fewer in high stakes domains like medicine, law, and finance when using web search. ...