OpenAI starts wider ChatGPT rollout of GPT-6 and interactive answers

Answers become something you can interact with

OpenAI announced a wider GPT-6 rollout for ChatGPT on October 7. Its new Intelligent UI can arrange an answer as text, visuals and controls rather than a single block of prose. The announcement describes a component library and a compiler that assembles those pieces as the model produces them, allowing the interface to appear progressively.

That changes the presentation, not the need to check the substance. Moving a slider or exploring a diagram may make an explanation easier to follow. It does not establish that the model's assumptions or information are correct. A polished answer can still need a source, a correction or a simpler explanation.

The plan and the reasoning mode both matter

The release notes say the global rollout began with Plus, Pro, Business and Enterprise in the Chat tab on October 7, with Free and Go due to follow from October 8. Enterprise access remains subject to administrator settings. A rollout start is not confirmation that every account already has the update.

Paid tiers use GPT-6 Sol for this experience; Free and Go use GPT-6 Luna. Intelligent UI supports reasoning settings from Instant through Extra High. The separate Pro reasoning option still uses GPT-6 Astra and does not support it. A Pro subscription and the Pro reasoning mode are therefore not the same thing.

OpenAI also says the models behind Work and Codex are not changing with this release. Its system card distinguishes the October Chat versions from earlier versions used elsewhere. The shared GPT-6 name should not be read as one identical update across every OpenAI product.

The safety report includes regressions

The October system card reports improved factuality on most of the company's difficult evaluation measures. It also records statistically significant regressions on some safety tests, including self-harm for both models and several categories in tests for younger users. OpenAI describes manual review and additional safeguards, including response blocking for certain sensitive teen content.

These results come from selected challenging cases, not a representative sample of everyday conversations. They cannot tell a reader how often a typical ChatGPT session will go wrong. Nor does the report establish that every aspect of safety improved.

For now, the useful distinction is between an interface that can help someone explore an answer and evidence that the answer is dependable. Availability on an individual account, the quality of generated controls and reliability outside the company's evaluations remain things to check, not promises we have verified.