OpenAI announced an update to GPT-5.3 Instant, ChatGPT’s most popular model, designed to make responses less “cringe” and reduce “preachy disclaimers.”

According to a blog post, the update focuses on improving accuracy and minimizing “unnecessary dead ends, caveats, and overly declarative phrasing” that can disrupt conversational flow.

The company also posted on X (formerly Twitter), showing a comparison between responses from GPT-5.2 Instant and the new GPT-5.3 Instant update.

Additionally, the updated model will pay more attention to tone, relevance, and conversational flow—factors that aren’t captured in benchmarks but can affect user experience.

OpenAI emphasised it will also reduce phrases like “Stop. Take a breath.”

OpenAI said that GPT-5.3 Instant reduces unnecessary refusals, softens “overly defensive or moralizing preambles,” and delivers answers without extra caveats.

The update also enhances answer quality by better balancing information from the internet with its own knowledge of the topic.