OpenAI发布GPT-5.3 Instant，高风险场景幻觉率降低约27%，默认替代5.2 Instant

According to 1M AI News monitoring, OpenAI has released GPT-5.3 Instant, replacing GPT-5.2 Instant as the default model for ChatGPT. This update focuses on improving daily conversation experience, mainly enhancing three aspects: reducing unnecessary refusals and preachy preambles, improving the quality of integrated web search answers, and lowering hallucination rates.

According to internal OpenAI evaluations, in high-risk scenarios such as medicine, law, and finance, hallucination rates when connected to the internet decreased by 26.8% compared to the previous version, and by 19.7% when relying solely on internal knowledge; in typical user-reported tricky dialogues, connected mode saw a 22.5% reduction, and offline mode a 9.6% reduction. OpenAI states that GPT-5.2 Instant’s tone was sometimes “cringe,” making inappropriate assumptions about user intent or making unnecessary statements, and version 5.3 has made targeted adjustments to address this.

The model is now available to all ChatGPT users, with the API provided as gpt-5.3-chat-latest. GPT-5.2 Instant will remain in the paid users’ legacy model list until it is retired after June 3.