OpenAI has released a limited version of its “Decisions API,” which makes quick judgments within a set of predefined options.

TechCrunch reported on Sept. 30 that the Decisions API is a tool intended to offset speed and cost constraints of large language models. It said the tool could be used for software automation and monitoring AI agents.

Sam Altman (샘 알트먼) explained at OpenAI’s annual developer event that the Decisions API works by giving OpenAI’s “Luna” model a predefined set of choices and having it pick from among them. That could include selecting an image classification category or choosing one among multiple agent actions. “If you narrow the range of choices, you can get very fast speed while still maintaining image understanding, broad language support and safety guardrails,” he said.

The Decisions API is structurally similar to “Zev,” released earlier this month by TypeSafe AI. Zev is a model for software automation that quickly and cheaply outputs probabilities for each option when developers provide multiple choices. Some developers are using it as an auxiliary model for large language models to increase speed and cut costs.

But it is not yet clear how similar the Decisions API is to Zev. OpenAI released it as a limited preview, and TechCrunch said there are not yet many developer cases that have verified its performance.

TypeSafe AI Chief Executive Diogo Almeida (디오고 알메이다) wrote on social media platform X (Twitter), joking about OpenAI’s move, calling it “the start of the clone wars.” He said OpenAI’s interest could be a signal that building in a way compatible with “System One” is the future.

TypeSafe AI has classified fast, intuitive judgment as System One and deliberative reasoning as System 2.

TechCrunch reported that concerns over speed and cost surrounding large language models are behind the attention on TypeSafe AI’s Zev. As criticism continues that directly applying large language models to much software is slow and expensive, it said Almeida is pointing to synthetic data that produces statistically meaningful outputs as the company’s competitive strength.