OpenAI has released GPT-5.3 Instant, an update to what it describes as the most-used model inside ChatGPT, with a clear focus on improving how everyday conversations feel and function.

The company says the new version is designed to make replies more fluid, relevant and better paced. It also claims improvements in accuracy, particularly when the model draws on web information, alongside a reduction in answers that end in refusals or extended caveats.

GPT-5.3 Instant is now available to all ChatGPT users. Developers can access it through the API under the name gpt-5.3-chat-latest.

According to the company, user feedback played a significant role in shaping the update. It said GPT-5.2 Instant sometimes declined questions that could have been answered safely and tended to adopt an overly cautious tone on sensitive subjects.

With GPT-5.3 Instant, the company says unnecessary refusals have been reduced. It also notes that the model has been adjusted to avoid defensive or moralising preambles, instead aiming to respond directly when a question is appropriate and safe to answer.

Tone was another area flagged by users. The previous version, the company acknowledged, could sound awkward or make assumptions about a user’s intent or emotional state. The new release is described as more concise, with fewer sweeping statements and a steadier personality across conversations and future updates.

Users retain control over tone settings, including warmth and enthusiasm.

The company also says the update changes how the model handles web-based responses. Rather than simply summarising search results, GPT-5.3 Instant is intended to prioritise key information earlier in its answers and avoid overemphasising links or loosely connected material.

On accuracy, the company cites two internal evaluations. One focused on higher-stakes areas such as medicine, law and finance. The other analysed de-identified ChatGPT conversations that users had flagged for factual errors.

In the higher-stakes review, the company reports a 26.8 percent reduction in hallucination rates when using the web and a 19.7 percent drop when relying on internal knowledge, compared with prior models. In the user-feedback evaluation, hallucinations fell by 22.5 percent with web use and 9.6 percent without it.

Beyond question-and-answer tasks, the company positions GPT-5.3 Instant as a stronger writing tool. It says the model produces more resonant and immersive prose and transitions more smoothly between practical tasks and expressive writing.

Some limitations remain. The company notes that responses in certain non-English languages, including Japanese and Korean, can still sound overly literal or stilted. It says feedback will continue to shape tone adjustments and expanded customisation options.

Updates to ChatGPT products labelled Thinking and Pro are expected soon. GPT-5.2 Instant will remain available for three months for paid users in a Legacy Models section.

Samsung Galaxy S26 Ultra Display Test Shows Slight Drop in Peak Brightness

The company says it conducted safety training and evaluations for GPT-5.3 Instant and has published further details in a system card, outlining its approach and findings. For now, the update reflects a measured attempt to refine how the model speaks, answers and balances caution with clarity.