樹林AI announces JJ-1, Japan's first AI model suite specialized in decision-making. JJ-1 is tailored to Japanese business workflows, cutting inference costs by up to 99% and accelerating processing.
樹林AI announced JJ-1 as Japan's first AI decision-making model suite. JJ-1 is designed to support decision-making processes in corporate operations and is tailored to Japanese business workflows.
JJ-1 was evaluated on a total of 6 datasets, combining public Japanese- and English-language datasets with proprietary data for Japanese business workflows, and recorded an overall accuracy rate of 79.20%. Its accuracy on proprietary data for Japanese business workflows was 97.43%.
JJ-1 cuts inference costs by up to 99%, enabling lower-cost inference than generative LLMs. Rather than generating output one token at a time, it scores predefined options to determine the next action.
In terms of processing performance, it can process typical requests in milliseconds and complete them in approximately 8 to 30 milliseconds on NVIDIA's B200. An official technical report summarizing the benchmarking methodology and cost comparison has been published, and a trial has also become available on a dedicated website.