GPT-6 in ChatGPT: Intelligent UI Arrives, Deprecation Looms

What happened

OpenAI began putting GPT-6 in front of ChatGPT users on 7 October 2026. Paying subscribers got GPT-6 Sol as the default model, and the free tiers followed a day later. 24 The company first introduced GPT-6 models in late September. This release is a set of new snapshots tuned for everyday conversation, replacing GPT-5.6 inside the consumer product. 24 One explainer dates the underlying models to 22 September and the interface rollout to 7–8 October. 1

The tier split is consistent across reports. Plus, Pro, Business and Enterprise subscribers get Sol, while Free and Go users get GPT-6 Luna starting 8 October. 4 Enterprise availability depends on workspace admins enabling it. 2 GPT-6 Astra reportedly remains the model behind Pro's higher effort level. 1

Intelligent UI is the real headline

The most visible change is a capability OpenAI calls Intelligent UI. Answers are no longer mostly prose. They can combine text, graphics and interactive elements in a single response. 24 Comparisons can appear side by side, explanations can become interactive charts, and users can ask for small working tools inside the chat, such as a savings calculator or even a simple game. 3 For more complex prompts, ChatGPT can build widgets that let users change inputs and watch the results update. 4

OpenAI says the model was trained to judge when an interface actually helps rather than adding one by default. A component library and compiler then render the output. 2 That design choice matters. It puts the decision about what an answer should look like in the model's hands, not the user's. Some will find that convenient. Others may see unwanted widgets where a sentence would have done.

The second change is speed. GPT-6 can start responding while it is still reasoning. 3 OpenAI says this cuts wait times by 44 percent and can improve answer quality, citing internal tests in which GPT-6 beat GPT-5.6 on difficult web searches. 3 Earlier partial answers are part of the same update. 2 These are company figures, and independent replication has not yet been reported.

The parts that don't line up

Coverage of features and tiers broadly agrees. The documentation does not. OpenAI's own announcement names GPT-6 Luna for the free tiers, yet the ChatGPT help centre reportedly says Free and Go users receive GPT-5.6 Luna instead. 1 Perhaps the help page simply lags behind the rollout. Still, conflicting statements about which model a user is getting make claims about free-tier capability harder to check.

The stranger signal concerns Sol itself. It has served paid ChatGPT seats since 7 October, yet OpenAI has reportedly removed it from its API pricing table, and an independent evaluator lists it as deprecated. 1 Earlier published pricing put Sol at $2.00 per million input tokens and $10.00 per million output tokens, and Luna at $0.10 and $0.50. 1 Why a model would be shown as deprecated in the developer channel while serving as ChatGPT's paid default is unexplained. One possible reading is that the consumer snapshots are being separated from the API versions. Another is that a successor is already planned. Neither is confirmed.

Benchmark positioning adds context. On one independent intelligence index, Sol scores 48 and Luna 38, while Anthropic's Claude Opus 5.5, dated the same day, scores 58. 1 If those numbers hold up, OpenAI's flagship consumer model does not lead on raw capability scores. That may help explain why the launch emphasizes interface and responsiveness rather than reasoning gains.

Why it matters

The scale is enormous. The rollout reaches a ChatGPT audience reported at more than 1.2 billion people. 1 Changing the default output format for that many users changes what people expect from a chatbot answer. Search engines, productivity tools and rival assistants are likely to feel pressure to offer similar interactive responses.

My reading is that this launch is a product-design bet more than a model-capability leap. Intelligent UI and faster partial responses are real, visible improvements. The inconsistent documentation and Sol's apparent deprecation in the developer channel, though, point to a release process running ahead of its own paperwork. Users should get clearer answers about which model is serving them, and developers should treat Sol's API status cautiously, before either group plans around it.