GPT-6 Release and Intelligent UI

GPT-6 introduces Intelligent UI for interactive responses

OpenAI has released GPT-6, introducing a new capability called Intelligent UI that allows ChatGPT to generate fully interactive user interfaces in response to user queries. This shift moves ChatGPT beyond a standard conversational format, enabling the model to compose responses using a combination of text, visuals, and interactive elements such as graphics, tappable buttons, forms, charts, and interactive diagrams based on the specific needs of the prompt.

Dynamic interface generation

Intelligent UI adapts the format of the response to the task. For example, comparisons are presented side-by-side, and complex explanations are delivered via interactive diagrams. The system can also create bespoke tools on the fly, such as bill splitters, savings calculators, or in-conversation games.

Technical implementation of Intelligent UI

Intelligent UI is powered by a library of native, streamable components and a compiler that processes the interface as the model generates it. This architecture allows the UI to appear progressively to the user without requiring the entire response to be completed first.

To enable this, OpenAI expanded training methods to evaluate the model's decisions regarding content, layout, and interactivity, teaching GPT-6 to determine when a complex interface is necessary and when a simple text response is sufficient.

Faster response times via interleaved thinking

GPT-6 improves the user experience by interleaving thinking with answering, allowing the model to begin responding while it continues to process complex problems. This eliminates the need for users to wait for a reasoning model to finish its entire thinking process before seeing the first token of the answer.

Performance benchmarks

  • Latency: In internal evaluations of high-value agentic tasks, GPT-6 Extra High begins answering in the same timeframe as GPT-5.6 Medium while achieving a higher overall score than GPT-5.6 Extra High.
  • Web Search: For queries requiring web search, GPT-6 Instant begins answering an average of 44% sooner than GPT-5.6 Instant.
  • Accuracy: GPT-6 demonstrates improved quality in web search results, making better decisions about when to search and more reliably finding supporting information. In internal evaluations of difficult problems, GPT-6 correctly addressed key aspects of the user's question more frequently than GPT-5.6.

Safety and alignment updates

GPT-6 incorporates safety advances from Astra to better resist jailbreak attempts, particularly multi-turn adversarial attacks. The model has been updated to strengthen protections against high-risk misuse in areas such as cyberattacks, biological threats, and violence.

Alignment evaluations focused on honesty and the model's ability to communicate its own limitations. GPT-6 is reported to be more capable of recognizing when it lacks the necessary information or tools to answer a prompt compared to GPT-5.6 Sol.

Availability and Model Variants

GPT-6 is rolling out globally across all ChatGPT tiers. The specific model powering the experience depends on the tier:

  • Plus, Pro, Business, and Enterprise: Powered by GPT-6 Sol.
  • Free and Go: Powered by GPT-6 Luna.

Rollout begins today for paid tiers and expands to Free and Go tiers tomorrow. This update applies specifically to the Chat experience; models powering Work and Codex remain unchanged.

Sources