OpenAI launches Decisions API with GPT-6 Luna for all customers

OpenAI officially launched its Decisions API for all customers on October 22, 2026, according to aitimes.com. The API is powered by GPT-6 Luna and is aimed at decision tasks that return probabilities, choices or scores.

OpenAI says the API can handle certain decision tasks up to 10 times faster than using GPT-6 Luna through the Responses API. Runtimes for tasks that previously took 1-3 seconds are reduced to about 100-180 milliseconds.

Developers define questions and allowed answers, then submit text, images or structured JSON data. The API supports native JSON input and questions containing images, and a single request can carry up to 400 questions, twice the previous limit. A setting lets customers process inference within the US and the EU.

Pricing is USD 0.10 per million input tokens and USD 0.01 for cached input, with no charge for output tokens. Eligible customers can use a Zero Data Retention option.

Early customers and speed gains

During the public beta, OpenAI added default integration with GPT-Live and prompt caching. The company uses the Decisions API internally to speed up responses from its voice assistant Dots, and early trial customers included Box, Datadog and Ramp.

Speak CTO Andrew Shee said that in a test integrating the Decisions API with GPT-Live, the time to next action dropped from 1.4 seconds to 0.13 seconds. OpenAI built the product by optimizing the inference process of GPT-6 Luna rather than training a separate decision-specific model, after looking at competitors' decision models.