The new API offers faster AI responses by narrowing focus to predefined choices, enhancing real-time applications.
The Decisions API, powered by GPT-6 Luna, is designed to perform tasks in a fraction of a second, significantly enhancing responsiveness.
It achieves this speed by narrowing the model's focus to a small set of predefined choices, maintaining image understanding, language support, and safety protocols while operating nearly ten times faster than previous models.
The Decisions API can automatically extract specific form fields from unstructured text inputs, streamlining data processing.
This capability enables automated routing of support tickets and inbound sales requests to the appropriate teams, with server-side response times observed at less than 100 milliseconds.
In gaming scenarios, the API processes camera frames to make split-second steering decisions, quickly choosing whether a car should stay in its lane or move left or right.
This speed makes the API suitable for computer use cases involving rapid actions, such as taking screenshots and immediately acting upon the visual data, operating significantly faster than standard reasoning models for specific decision tasks.
Integrating GPT-Live-1 with the Decisions API allows for more responsive live interactions, where GPT-Live-1 handles voice synthesis.
Simultaneously, the Decisions API selects appropriate character expressions in real-time as the user speaks, making digital assistants feel more natural and engaging.
The Decisions API enables a robot to use camera input to identify objects and direct its head movement, handling ambiguous requests like 'follow the fruit' or comparing objects.
The API demonstrates the ability to effectively distinguish between objects, such as a gamepad and an apple, based on contextual understanding.
The robot uses its camera input and the Decisions API to determine where to look, allowing it to identify and track objects like an apple based on voice commands.
Target identification is achieved through the Decisions API, which analyzes camera frames to interpret conceptual commands like 'follow the fruit' without requiring specific object names.
By analyzing camera frames periodically, the Decisions API identifies the correct target, showing the robot's ability to adapt to complex prompts by conceptualizing which items fit a given category.
In fact, if you want to try with a different request that's a bit more ambiguous, feel free.
When presented with an apple and a gamepad and asked, 'Lavender, what's more fun to play with?', the robot chose the gamepad, demonstrating its ability to process subjective reasoning.
This shows the API's capability to maintain focus and track a chosen object, even when its position is manually shifted by the user, adapting to moving environmental inputs.
The demonstrated applications of the Decisions API combined text and vision inputs, supporting various trigger methods including keyboard and voice interactions, with OpenAI aiming to enable a new category of AI interaction.
Answers come from the transcript, with the exact spot cited.
Want the next article from OpenAI?
When OpenAI publishes, we'll write it up like the one you just read and email it to you.
OpenAI published 4 in the last 7 days.
