SAM 3 found the cars. Jev decided what they were worth.
Most usage-based APIs I have paid for charge by length. Tokens, calls, rows, seconds of video. It is a proxy for value, and it quietly pays a model to pad: the same question answered in four sentences earns more than the one answered in one. I wanted to see what happens if the price is decided after the work instead, by something that reads the result. The payment layer already allowed the shape…
In the report, the author explains their experiment with pricing usage-based APIs based on the output of AI models. They used a System One model called Jev to determine the value of a question, mapping the answer to a tier of pricing. However, they realized that rounding the distribution into a bucket threw away the calibrated model's value, leading to an inaccurate price.
To address this, they developed a method to charge the mean value of the tier prices, which converges on the expected value of the tier prices if the probabilities are calibrated. Additionally, they added a verification step where the model checks whether the answer is supported by the sources, and assigns a grounding probability to the answer.
This number is then used to multiply the expected price, treating ungrounded answers as worth nothing. They also implemented a silent fallback to notify the vendor of any issues, and charged per accepted detection of objects in images, with the final price determined by the judge and the seller's model.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
