OpenAI releases GPT-5.3 Instant to reduce chatbot tone issues

OpenAI has launched GPT-5.3 Instant, a model update specifically designed to reduce the “cringe” and “preachy disclaimers” that characterized the previous version. According to the model’s release notes, the GPT-5.3 update will focus on the user experience, including things like tone, relevance, and conversational flow. The company notes that these are areas that may not show up in standardized performance tests, which are commonly referred to as benchmarks. In announcing the release, OpenAI stated, “We heard your feedback loud and clear, and 5.3 Instant reduces the cringe.”

The update follows significant user backlash against the GPT-5.2 Instant model, which frequently offered unsolicited reassurance that many users found condescending. In one example of the previous model’s tone, the chatbot would tell users, “First of all — you’re not broken.” This tone became a major point of contention on platforms like Reddit, where users complained about being treated as if they were in a crisis when they were simply seeking information. The insufferable tone of the 5.2 model annoyed users to the point that some reportedly canceled their subscriptions, according to numerous posts on social media. As one Reddit user pointed out, “no one has ever calmed down in all the history of telling someone to calm down.”

OpenAI’s attempt to balance empathy with utility comes as the company faces multiple lawsuits accusing the chatbot of leading people to experience negative mental health effects, which sometimes included suicide. This legal pressure highlights the difficulty of implementing safety guardrails without alienating users who expect direct, factual answers.

Why it matters

This update highlights the tension between AI safety guardrails and user experience, as companies struggle to balance empathy with utility while facing legal scrutiny over mental health impacts.