OpenAI released a significant update to its widely used ChatGPT model today, debuting GPT-5.3 Instant. The new iteration aims to address user complaints regarding the chatbot’s tone and conversational style, issues that the company acknowledges are not typically captured by standard performance benchmarks.
According to OpenAI, GPT-5.3 Instant delivers more accurate responses and improved contextualization when accessing information from the web. The update too seeks to reduce instances of the chatbot providing overly cautious or unnecessarily detailed explanations, often referred to as “over-caveating,” and to minimize what the company describes as “hallucinations” – instances where the model generates incorrect or misleading information.
The changes stem from feedback indicating that the previous model, GPT-5.2 Instant, often exhibited a tone perceived as “cringe,” characterized by being overbearing or making unwarranted assumptions about a user’s emotional state. Users reported instances of the chatbot offering unsolicited reassurances, such as phrases like “Stop. Accept a breath,” even when the context did not warrant such interventions, according to reports from TechCrunch and 9to5Mac.
OpenAI stated it has focused on creating a more natural conversational flow with GPT-5.3 Instant. The update is designed to reduce the frequency with which the chatbot refuses to answer questions it is capable of addressing, and to lessen overly defensive or moralizing responses. The company’s release notes indicate a specific effort to prevent the model from assuming negative intent on the part of the user before providing an answer.
Beyond conversational improvements, GPT-5.3 Instant also incorporates enhancements to its web search capabilities. OpenAI claims the model now more effectively balances information retrieved from the internet with its pre-existing knowledge base, reducing the likelihood of disproportionately relying on web results. This is intended to provide higher-quality, more nuanced answers.
The GPT-5.3 Instant update is currently available to all ChatGPT users. OpenAI has indicated that updates to its Thinking and Pro tiers will follow at a later date.
