OpenAI makes Decisions API available to all customers, stepping into the decision model race

The fast, low-cost AI classification market, pioneered by $7.6 billion startup TypeSafe, now faces a new contender

By LineZotpaper

Published22 October 2026

Read Time3 min

Reactions1 on X

OpenAI has opened its Decisions API to all customers, entering a market for fast, cheap AI models that make split-second choices, a category largely created by rival startup TypeSafe and its Jev model. The move pits a repurposed version of OpenAI's ChatGPT Luna against a wave of newcomers, some of which already outrank it on public benchmarks.

OpenAI has made its Decisions API available to all customers, its first wide release in a fast-growing market for small, low-cost AI models built to make quick choices rather than generate long text.

TypeSafe, a startup founded by a former OpenAI researcher, effectively created the category with its Jev model, and the niche helped make the company worth $7.6 billion. These "decision" or "System One" models act as super classifiers: given a set of options, they return a choice and a confidence rating in milliseconds, at a fraction of the cost of full-sized large language models. Developers have found they can replace LLMs in many workflows with faster, cheaper results.

Nikunj Handa, product lead for OpenAI's APIs, said his colleagues saw Jev and reacted with, "wow, we have like a million use cases for this." Rather than build a specifically calibrated decision model, the team took ChatGPT Luna and reconfigured it to speed up inference and cut cost. "The big unlock with Decisions API has been the speed, shrinking things from one to three seconds to 100, 120, 180 milliseconds," Handa said. "Right now, it's just plain old Luna that gets the job done."

Announced at OpenAI's developer day on September 29, the API became generally available this week, adding voice and visual modalities and targeting inference resources geographically to lower latency around the world. Early trials with customers including Box, Datadog, Ramp and Speak used the API to replace LLMs with higher throughput. Andrew Hsu, chief technology officer at AI language tutor company Speak, said the API cut the time between actions from 1.4 seconds to 0.13 seconds.

Competition is fierce. Public benchmarks such as the AIM System One Index and the Artificial Analysis Decision Index currently favor D1 Ultra, a decision model Liquid released last week, and Base Lab's Elephant-1. The latest Jev model ranks seventh and third on those indices, while OpenAI's model sits eighth and fourteenth. Still, TypeSafe's offerings remain the most widely queried on the OpenRouter routing service.

TypeSafe chief executive Diogo Almeida distrusts benchmarks and urges customers to trial models in their own work. The most playful test, JevMan, has decision models play PacMan, and the latest Jev model tops that ranking. Felix Wunderlich, chief growth officer at Opper AI, which built the benchmark, said PacMan "is an interesting test because of its nonstop stream of quick, high-stakes decisions."

OpenAI's argued advantage is multimodality. Handa said Luna "is extremely powerful and has been showing better calibrations" on image classification, and customers want image outputs as well. He sees further speed gains ahead, asking whether responses could reach 60 milliseconds, which would allow an API call for every frame of a video.

The model's namesake is the economist William Stanley Jevons, who observed that using a resource more efficiently tends to make it more widely used. Handa expects an expanding "crazy set of use cases" that were hiding in plain sight until decision models revealed them.

Analysis

Why This Matters

Decision models open a new price-performance tier in AI: fast, cheap classification that can replace large language model calls in many workflows, cutting both cost and latency.
OpenAI's entry validates a category a startup created and sharpens competition at the model level, which tends to push prices down for developers.
The outcome will shape which workloads move off heavy LLMs and what developers can build in real time, from responsive voice assistants to video-frame processing.

Background

Decision models, sometimes called System One models, are a narrower tool than general-purpose chatbots: they take a set of options and return a choice with a confidence score, trading open-ended reasoning for speed and low cost. Where a large language model might take seconds to answer, a decision model can respond in milliseconds, making it attractive for high-volume automation, routing, moderation and interactive voice applications. The category's name nods to the Jevons paradox, the 19th century economist William Stanley Jevons's observation that making a resource cheaper and more efficient tends to increase, not reduce, its overall use.

Key Perspectives

OpenAI: The API team moved quickly to follow the category leader, repurposing its existing Luna model rather than training a bespoke decision model, and betting that speed plus multimodal support will win customers. Handa says plain-old Luna "gets the job done" for now.

TypeSafe: The startup created the category and still leads in real-world usage on OpenRouter. Chief executive Diogo Almeida warns that benchmarks are unreliable and wants customers to test models in their own workflows, while continuing to ship features such as image support and "smart-if" statements.

Rivals and benchmark watchers: Liquid's D1 Ultra and Base Lab's Elephant-1 currently top public indices, showing the lead is contested. The gap between benchmark rankings and actual query volumes suggests the market has not settled on a clear winner.

Developers and customers: Early users such as Speak report dramatic latency gains, suggesting the models deliver practical benefits beyond marketing claims, though what makes a good decision model seems to vary by workload.

What to Watch

Whether OpenAI reaches its stated target of 60 millisecond responses, which would enable per-frame video processing and other new applications.
How closely public benchmark rankings track real adoption as more developers run trials, and whether TypeSafe retains its lead in query volume.
The next Jev release from TypeSafe, which tops the PacMan benchmark and has been expanding into image inputs, and the response from Liquid and Base Lab.

Sources

OpenAI makes its Jev competitor widely available as decision model competition heats up · AI News & Artificial Intelligence | TechCrunch

Checked

10 claims in this story checked against 1 source: 6 supported.

Show each claim

SupportedOpenAI made its Decisions API available to all customers and entered the decision-model market.AI News & Artificial Intelligence | TechCrunch: “OpenAI made its Decisions API model widely available to users for the first time, entering the market for fast, cheap intelligence”
SupportedTypeSafe pioneered the category, and the startup is valued at $7.6 billion.AI News & Artificial Intelligence | TechCrunch: “TypeSafe’s Jev model created a category for deep learning models”
SupportedThe decision-model category was largely created by TypeSafe and its Jev model.AI News & Artificial Intelligence | TechCrunch: “TypeSafe’s Jev model created a category for deep learning models”
Not in the sourcesTypeSafe was founded by a former OpenAI researcher.AI News & Artificial Intelligence | TechCrunch: “a startup founded by a former researcher at the frontier lab”
Partly supportedHanda’s colleagues reacted to Jev with, “wow, we have like a million use cases for this.”AI News & Artificial Intelligence | TechCrunch: ““wow, we we have like a million use cases for this.””
Partly supportedDecision models return a choice and confidence rating in milliseconds.AI News & Artificial Intelligence | TechCrunch: “Given a set of options, they can quickly make choices and offer a confidence rating in their decision.”
SupportedOpenAI reconfigured ChatGPT Luna to speed up inference and lower costs rather than build a specifically calibrated decision model.AI News & Artificial Intelligence | TechCrunch: “Rather than building a specifically calibrated decision model, the team took ChatGPT Luna and reconfigured it to speed up the inference and lower the cost.”
SupportedSpeak said the API reduced the time between actions from 1.4 seconds to 0.13 seconds.AI News & Artificial Intelligence | TechCrunch: “Andrew Hsu, the CTO at Speak, which builds AI language tutors, said the Decisions API allowed the company to cut the time between actions from 1.4 seconds to .13 seconds.”
SupportedThe latest Jev model ranks seventh and third on the two named indices, while OpenAI’s model ranks eighth and fourteenth.AI News & Artificial Intelligence | TechCrunch: “The latest Jev model remains competitive on both indices, in seventh and third, with OpenAI’s Decision Model falling eighth and fourteenth, respectively.”
Partly supportedThe model’s namesake is the economist William Stanley Jevons.AI News & Artificial Intelligence | TechCrunch: “they live up to Jev’s namesake, an economist who explained how using a resource more efficiently makes it more ubiquitous.”