GPT-5.3 Unveiled: OpenAI Promises a Less Cringe, Less Preachy ChatGPT

On Tuesday, March 3, 2026, OpenAI rolled out GPT-5.3 Instant, a fresh iteration of its popular conversational model. This update focuses not on raw performance enhancements but on improving the daily user experience. Key areas of improvement include the tone of responses, the relevance of the information provided, and a reduction in the occurrence of fabricated responses, known as “hallucinations.” This update has become a priority following weeks of harsh criticism from users.

OpenAI Adjusts ChatGPT’s Tone Following Fresh User Backlash

The previous model, GPT-5.2 Instant, was widely criticized by users in recent weeks for its condescending tone. Many shared their frustrations on Reddit and other social media platforms, where they reported instances of ChatGPT adopting a paternalistic tone, asking users to calm down or reassuring them unnecessarily, even on straightforward factual inquiries, sometimes suggesting they “take a deep breath.” This behavior, described as “cringe” by the community, led some subscribers to consider canceling their subscriptions.

With GPT-5.3 Instant, OpenAI aims to address these issues. On X, the company stated: “We’ve heard your feedback loud and clear, 5.3 Instant reduces the cringe.” In its blog post, OpenAI detailed that the update focuses on “tone, relevance, and the smoothness of interactions,” aspects that do not typically appear in benchmarks but significantly affect daily user interactions.

The new model aims to eliminate moralizing preambles and the refusal to answer legitimate questions. Now, when a user posts a valid query, GPT-5.3 Instant is designed to respond directly without unnecessary warnings.

Fewer Hallucinations and Improved Web Searching

Beyond just the tone, OpenAI claims measurable improvements in reliability. According to the company’s internal assessments, the rate of hallucinations has decreased by 26.8% when the model utilizes web search, and by 19.7% without web access. For conversations flagged by users for factual errors, the reduction rates are 22.5% and 9.6%, respectively.

Another area of enhancement is in handling search results. The model now better contextualizes information found online, rather than merely listing links, and balances web sources more effectively with its own knowledge.

GPT-5.3 Instant is now available to all users of ChatGPT and to developers through the API. The Thinking and Pro versions will be updated soon. The previous model, GPT-5.2 Instant, will remain available to paying subscribers for three months, with a planned discontinuation on June 3, 2026.