OpenAI Ships GPT-5.3 Instant to Fix the Tone

OpenAI rolled out GPT-5.3 Instant as the new default in ChatGPT, a model update built almost entirely around making the assistant less preachy and more direct. Hallucination rates dropped between 10% and 27% depending on the evaluation, and the company retired the safety-disclaimer-first pattern that made GPT-5.2 feel like talking to a compliance officer. The “more accurate, less cringe” framing is unusually candid – a tacit admission that last year’s GPT-5 launch, which triggered a wave of cancellation threats over its colder tone, left real damage. GPT-5.2 Instant gets a three-month sunset ending June 3. Every major lab is now iterating on personality and refusal calibration as aggressively as they once chased benchmark scores.