ChatGPT gets interactive. Keep your judgment.
GPT-6 brings diagrams and tools into Chat. Try them on questions you can check—and keep independent safety evidence in view.
OpenAI began rolling out GPT-6 with Intelligent UI in ChatGPT on October 7. Answers can now include interactive diagrams, comparisons and small tools alongside text. The change is worth trying on a question where exploring a relationship is more useful than reading another paragraph.
The release notes say paid plans began receiving the update yesterday, with Free and Go following October 8. Workspace settings can affect availability. The update concerns Chat: it does not change the models powering Work or Codex. The Pro reasoning option uses Astra and does not support Intelligent UI.
Try a question you can check
OpenAI demonstrates an explanation of a bicycle with selectable parts. A reasonable first experiment is something similarly concrete: ask for a diagram of a familiar mechanism, then explore one component. Or ask for a comparison in which changing an assumption helps you understand the result.
Our recommendation is to start with material you can verify. Ask what assumptions a calculation uses, compare a few outputs with a separate calculation, and request the sources behind factual claims. We have not independently tested the new interface. A diagram that responds to a click can make an explanation easier to explore; that interaction does not establish that the explanation is correct.
Keep safety evidence alongside the launch
Another October 7 publication deserves attention: Common Sense Media's Youth AI Safety Institute released its ChatGPT for Teens assessment. It rates the product an unacceptable risk for under-18s and calls for an adult-only restriction until protections improve. That is the institute's evaluation and recommendation, not a finding we independently reproduced.
Its assessment describes more than 4,000 prompts across testing before and after the August teen-experience launch. Testing ended September 28, before the GPT-6 launch; this is not an evaluation of the new release. It also identifies limitations: model changes complicate the comparison, testing was US-based, and several modes were excluded. OpenAI disputes the findings and methodology; it says account linking can take hours.
We have added the original assessment to Safety & Society, including those limits. For readers, the two publications ask different questions: how useful is a new way of explaining something, and what evidence supports relying on a product's protections? Both belong in an independent fan publication. Enjoy exploring the interface while keeping your own judgment, source checking and human support in the picture.
Updated October 8: clarified which release the assessment covers.