OpenAI Opens Its Decisions API to Everyone, Chasing the Fast Cheap Decision Model Race

OpenAI just picked a fight with a $7.6 billion startup by copying its best trick.

On Thursday, the company opened its Decisions API to all customers, a new endpoint that doesn’t write paragraphs or hold conversations. It makes split-second structured calls: is this product photo damaged, should this ticket get escalated, which option should the agent pick next. The price starts at $0.10 per million input tokens, with output free.

The launch is OpenAI’s first wide release in the fast-growing “decision model” market, a niche largely created by rival startup TypeSafe and its Jev model. OpenAI’s version runs on a retuned GPT-6 Luna and answers in as little as 100 milliseconds, roughly ten times faster than going through its standard Responses API.

What the Decisions API actually does

Most AI APIs hand you a wall of text and leave the parsing to your code. The Decisions API skips that entirely. Developers send evidence (a string of text, an image, or both) plus an array of named questions, and the model answers each one in a fixed, machine-readable shape.

There are three answer types. Predicate questions come back as a probability between 0 and 1 that a condition holds, like whether a delivery photo shows a visible package. Choice questions return one of the developer’s own predefined values, with probabilities and a confidence figure attached. Score questions grade input against an ordered rubric of 2 to 10 levels and return a probability-weighted average. A result can land between two levels, rather than snapping to one.

No essays. No pleasantries. Just a number.

Requests hit a dedicated endpoint, POST /v1/decisions, and gpt-6-luna is the only model it supports for now. OpenAI first teased the tool at its Developer Day on September 29, ran a public beta starting October 6, and now calls it generally available.

Speed is the whole pitch

Nikunj Handa, OpenAI’s head of API products, told TechCrunch that in some tasks the team cut latency from one to three seconds down to 100, 120 or 180 milliseconds. “The biggest breakthrough of the Decisions API is speed,” he said, adding that the team retuned the existing ChatGPT Luna for inference speed and cost rather than training a dedicated decision model. “Right now, it’s just plain old Luna that gets the job done.”

Early testers say the numbers hold up. Speak, which builds AI tutors for language learners, told TechCrunch its engineers cut the time between actions from 1.4 seconds to 0.13 seconds by swapping OpenAI’s larger models for the new endpoint. Box, Datadog and Ramp also ran early trials, and OpenAI says it used the API itself to speed up the responses of its Dots voice assistant.

The public release adds voice and visual capabilities, and OpenAI has spread inference capacity across regions to cut latency worldwide.

A $7.6 billion startup lit the fuse

The backstory is unusually blunt for OpenAI. TypeSafe, founded by a former Frontier Labs researcher, built Jev, a “super classifier” that picks options from a set and returns a confidence score in milliseconds, at a fraction of an LLM’s cost. Developers pounced on it for agent workflows, and the startup is now worth $7.6 billion, CoinDesk reported.

Handa didn’t hide where the idea came from. When his colleagues saw Jev, he told TechCrunch, the reaction was, “Wow, we have about a million use cases.” Rather than build a purpose-built, calibrated decision model from scratch, his team took ChatGPT Luna and reconfigured it to run faster and cheaper.

That framing puts OpenAI in the role of fast follower, racing a startup it dwarfs in every dimension but one: the new category’s home turf.

The bigger picture

There’s a structural shift hiding behind the pricing table. AI agents make hundreds of small branching decisions per workflow: route this, score that, escalate or ignore. Every one of those steps currently burns a full LLM call. When tokens cost dollars per million and responses take seconds, the agent economy strains. A model that answers in 120 milliseconds for ten cents per million inputs makes always-on agents cheap enough to be boring.

That is also why competitors are flooding in. A wave of small “decision models” has hit the market in recent weeks, some already outranking OpenAI’s entry on public benchmarks, according to coverage of the space. OpenAI’s advantage is distribution: millions of existing API customers can now flip a switch. But the category’s economics reward whoever is cheapest and fastest, not whoever is biggest.

There are gaps to watch. OpenAI published no accuracy figures with the launch, and the dataist.ai coverage of the beta noted that a clean yes-or-no output is useful only if teams can judge when to trust it. For compliance-heavy jobs like content moderation or insurance triage, confidence scores are only half the story; calibration is the other half. OpenAI says it may do specialized training for decision tasks later. For now, speed is doing the selling.

Why this matters

OpenAI has spent two years selling bigger models for bigger problems. The Decisions API sells something smaller for smaller problems, and that may be the smarter business. Every agent framework on the market has a routing layer, and OpenAI just offered to own it for a dime per million tokens. If developers take the offer, the humble decision call becomes the stickiest piece of OpenAI’s platform: not the flagship model everyone talks about, but the plumbing every agent quietly runs on.

FAQ

What is OpenAI’s Decisions API? A new endpoint that returns structured decisions — a probability, a choice, or a score — instead of free-form text. It runs on GPT-6 Luna at $0.10 per million input tokens.

How is a decision model different from a regular LLM API? Decision models answer only bounded questions and return machine-readable results in milliseconds for a fraction of the cost. They handle the branching steps inside AI agent workflows, not text generation.

Why did OpenAI build this now? Startup TypeSafe created the category with its Jev model and is now worth $7.6 billion. OpenAI API chief Nikunj Handa told TechCrunch his team saw Jev and realized they had “like a million use cases for this.”

What does the Decisions API cost? $0.10 per million input tokens, $0.01 per million for cached input, no charge for output tokens. Cache writes are billed separately, and eligible customers get Zero Data Retention and HIPAA-compliant processing.

Sources: TechCrunch, CoinDesk, WebProNews, Pandromeda.