OpenAI has made its decision model widely available for the first time, opening the OpenAI decision model to all customers through its Decisions API. The launch puts OpenAI in direct competition with TypeSafe's Jev, a fast model that picks between options and gives a confidence rating for each choice. The Decisions API was announced at the company's developer day on September 29 and became available to all customers on Thursday, 22 October 2026. Whether it can match Jev is less clear, because public rankings still place TypeSafe's latest model ahead.
Timeline: what has happened so far
- September 29: OpenAI announced the Decisions API at its developer day.
- The week before the release: TypeSafe released its own rankings alongside a new Jev model that supports images and smart-if statements.
- Thursday, 22 October 2026: The Decisions API became available to all customers.
How the decision model works
Decision models are deep learning systems that act as super classifiers. Given a set of options, they choose quickly and attach a confidence rating to the choice. TypeSafe's Jev created this category, and software developers soon found that such models could replace large language models, the text-generating systems behind chatbots, in many workflows at higher speed and lower cost.
OpenAI did not train a new, specially calibrated model for this. Instead, its team took ChatGPT Luna and reconfigured it to speed up inference and reduce cost. Nikunj Handa, product lead for OpenAI's APIs, said the speed was the main gain, with some tasks falling from one to three seconds to between 100 and 180 milliseconds. He added that the company may add specific training later, but for now the existing Luna model does the job.
In early trials, customers including Box, Datadog, Ramp and Speak used the API in place of OpenAI's larger language models. Andrew Hsu, chief technology officer at Speak, which builds AI language tutors, said the change cut the time between actions from 1.4 seconds to 0.13 seconds. OpenAI also used the model to make its Dots voice assistant more responsive. The public version now adds voice and visual options, and the company said it has placed its computing resources in different regions to reduce delays.
Benchmarks and competition
It is not settled what makes a good decision model. TypeSafe's chief executive, Diogo Almeida, distrusts benchmarks and urges users to test models on their own work. On the AIM System One Index and the Artificial Analysis Decision Index, the latest Jev ranks seventh and third. OpenAI's decision model ranks eighth and fourteenth on the same two indexes, and the benchmarks favour D1 Ultra from Liquid and Elephant-1 from Base Lab. TypeSafe's models are also the most queried on OpenRouter, a platform that routes requests to AI models.
In JevMan, a benchmark built by Opper AI in which decision models play PacMan, the latest Jev model tops the ranking. Felix Wunderlich, the company's chief growth officer, said the game tests a model's ability to make many fast, high-stakes choices. OpenAI may have an edge in working with images, and Handa said its model has shown better calibration on image classification. He also said that much faster responses, around 60 milliseconds, could allow apps to call the model for every frame of a video.
Aasan alfaaz mein
OpenAI ne apna decision model Decisions API ke zariye sab customers ke liye khol diya hai. Yeh model fori faislay tez aur kam kharch mein karne ke liye bana hai, aur kuch kaamon mein jawab 100 se 180 millisecond tak aa sakta hai. Benchmarks mein TypeSafe ka Jev abhi aage hai, is liye muqable ki race abhi khuli hai. Aap ke liye iska matlab yeh hai ke aage chal kar aap ki AI apps zyada tez jawab de sakti hain.
