OpenAI Decisions API launches finite-choice inference

// 1h agoPRODUCT LAUNCH

OpenAI Decisions API launches finite-choice inference

OpenAI’s Decisions API focuses Luna on user-defined questions with finite answers, returning choices for classification, request routing, and selecting an agent’s next action. It enters limited preview with text and image inputs, with broader availability planned soon.

// ANALYSIS

The API’s real appeal is control: developers get model-powered judgment without handing an agent unlimited freedom.

–Finite outputs are easier to validate, log, evaluate, and integrate than free-form responses
–High-volume routing and triage could benefit from lower-latency, narrower inference
–Developers still need deterministic policy checks and human escalation for consequential decisions
–The limited preview leaves reliability, calibration, pricing, and latency benchmarks unanswered
–OpenAI is formalizing a new layer between traditional classifiers and open-ended agents

// TAGS

decisions-apiapiinferencestructured-outputagenttool-useautomation

DISCOVERED

1h ago

2026-09-29

PUBLISHED

1h ago

2026-09-29

RELEVANCE

9/ 10

AUTHOR

OpenAI

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

UPDATE30m ago

Scenario Turns Images Into Editable 3D Scenes

Scenario now turns a single image into a game-ready, editable 3D scene with one click, powered by Claude Opus 5.5. The workflow runs natively in Scenario without MCP or a Claude subscription.

UPDATE38m ago

fal Agent opens creative workspace to everyone

fal Agent is now broadly available as a pay-as-you-go creative workspace for generating and editing image, video, audio, and 3D assets. Persistent project memory, model selection, voice interaction, batch workflows, and stackable Skills aim to turn fal’s model marketplace into an end-to-end production environment.

UPDATE40m ago

Grok Bot Gets Custom AI Avatars

Grok Bot now lets users replace its default avatar with an uploaded image or a generated picture created from a short prompt. Each Bot can have its own visual identity alongside its name, description, and working context.