OpenAI launches GPT-5.3 Instant, an update to widely used model

Today, OpenAI rolled out GPT-5.3 Instant, an update to its most widely used model powering ChatGPT. The new version focuses on real-world usability rather than benchmark scores, delivering fluid, accurate, and natural conversations with less robotic tone and more helpful information.

GPT-5.3 Instant resolves issues like overlay cautious refusals, lengthy disclaimers, and stiff phrasing. It reduces unnecessary safety warnings, cuts defensive or moralizing language, and provides direct answers when needed. It also improved web search integration. Instead of dumping long link lists, the model surfaces relevant information upfront with the latest online data.

Benchmark data shows that its hallucinations dropped by 26.8% when using web access and 19.7% without. User-flagged errors fell by 22.5% with web and 9.6% without. Creative writing now has richer prose, better texture, and more immersive storytelling.

Side-by-side comparisons shared by OpenAI confirm the difference. Where GPT-5.2 Instant refused a simple archery trajectory question by showing safety guideline restrictions, GPT-5.3 Instant immediately models projectile motion, range, and drag effects. For live sports news, it notes Kyle Tucker’s $240 million Dodgers deal and CBA implications instead of outdated information.

OpenAI GPT 5.3 Instant

The model is now available to all ChatGPT users and developers through the API as “gpt-5.3-chat-latest.” GPT-5.2 Instant remains accessible for paid users for three more months as a legacy option.

OpenAI has also confirmed that the team is working on non-English languages and tone consistency, but the release already marks a major step toward AI that feels like a reliable daily companion rather than a guarded assistant.

Furthermore, updates to Thinking and Pro modes are coming soon. GPT-5.3 Instant sets a new standard for accessible, human-like interaction.