AI BriefingOpenAI Releases Decisions API in Public Beta for Structured Judgment Tasks
Key point
The new endpoint returns structured probabilities for tasks like fraud detection and classification, charging only for input tokens.
Details
OpenAI has released the Decisions API in Public Beta, a new endpoint designed for scenarios requiring model judgment rather than text generation. Users can input data and ask specific questions—such as "Is this fraud?", "Which category does this belong to?", or "How severe is this?"—to receive structured probabilities instead of free-form paragraphs. This makes the API suitable for routing, classification, moderation, and automated workflows.
Pricing and Architecture
The API features a distinct pricing model compared to standard text generation: users pay for input tokens only, with no output-token charge. The service is backed by GPT-6 Luna, exposing decision-making capabilities through this specialized interface.
Comparison to TypeSafe’s Jev
The article notes similarities to TypeSafe’s Jev, which also focuses on making structured decisions from unstructured data. However, there are key differences:
- Jev is presented as a model built specifically for decision-making, with claims of very fast parallel inference and well-calibrated probabilities.
- OpenAI’s Decisions API relies on GPT-6 Luna, and it remains unclear whether its internal mechanics or probability calibration match Jev’s claims.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Benchmark Shows GPT-6 Luna's Confidence Scores Unreliable for Complex DecisionsOpenAI Launches $500 ChatGPT Pro Tier with Exclusive Access to Astra Ultrafast