OpenAI makes its fast Decisions API open to all customers
Speak, a firm that makes AI language tutors, cut the time between steps from 1.4 seconds to 0.13 seconds with the API.
On Thursday OpenAI made its Decisions API, a model that picks one alternative from a set, available to all customers. OpenAI made it from its ChatGPT Luna model to compete with Jev from TypeSafe, a startup with a value of $7.6 billion. OpenAI says the API cuts the time for a result from one to three seconds to 100 to 180 milliseconds.
How it works
OpenAI did not train a new model. It changed Luna to operate at a higher speed and a lower cost. The model gives a confidence rating with each result, and the version for all customers adds voice and visual input.
The numbers
Speak and other customers tried the API first. Speak says the time between steps fell from 1.4 seconds to 0.13 seconds. On two rankings, the OpenAI model is eighth and fourteenth, and Jev has a position above it on each.
What is not known
The source of a good decision model is not known, other than that it works for you. The correct alternative is different for each task. The CEO of TypeSafe tells users to test models on their work.
Insight: Large labs now chase a category that a startup made
OpenAI made its API from Luna, and its model is eighth and fourteenth on two rankings. Microsoft says its Decision-1 model is 2.8 times as fast as Jev in its latency test. This reads as a race: large labs copy the category of a startup and compete on speed and price. No rankings can replace a test on your work.
Sources
Posted
