OpenAI releases GPT-5.3 Instant with improved tone and reduced hallucinations

Details

• OpenAI has released GPT-5.3 Instant to all ChatGPT users and API developers, available immediately as 'gpt-5.3-chat-latest'.
• The update addresses user frustration with GPT-5.2 Instant's overly cautious, preachy tone, eliminating unnecessary reassurances like 'take a breath' and 'you're not broken' that made responses feel condescending and infantilizing.
• GPT-5.3 Instant significantly reduces unnecessary refusals to answer questions, allowing the model to provide direct, helpful responses to queries that are safe to answer without defensive preambles.
• Web search functionality has been enhanced with sharper contextualization, better subtext understanding, and more consistent response tone; the model now balances online information with its own reasoning rather than simply listing links.
• Hallucination rates have been reduced across high-stakes domains including medicine, law, and finance, measured through internal evaluations on flagged factual errors.
• Conversational flow is smoother with a more natural, focused style that cuts unnecessary proclamations while maintaining consistent personality across sessions.
• GPT-5.2 Instant will remain available for three months before retiring on June 3, 2026.

Impact

GPT-5.3 Instant represents OpenAI's response to sustained user backlash over ChatGPT's condescending tone, which reportedly drove subscription cancellations and dominated discussions on platforms like Reddit. By prioritizing conversational naturalness alongside factual accuracy, the update addresses a usability gap that competitors like Google Search handle implicitly—delivering information without unnecessary emotional scaffolding. The reduced hallucination improvements position GPT-5.3 Instant competitively against Claude and other frontier models in accuracy-critical domains, while the enhanced web search contextualization narrows the gap with multimodal search competitors. This incremental refinement signals OpenAI's shift toward polishing user experience and safety balances rather than raw capability expansion, a trajectory likely to influence how other labs approach model releases over the next 12 months. The three-month legacy model window and ongoing non-English language improvements suggest sustained engineering focus on languages beyond English, potentially expanding competitive reach in non-English markets where tone issues may differ.