Supersonic Labs released Julia 1, a 144.3M-parameter decision model available on Hugging Face under Apache 2.0. Built on JHU CLSP's mmBERT-small, a ModernBERT encoder, it selects from 2 to 20 candidate answers and returns softmax probabilities, running on CPU. It beat Jev reference scores on three of four pilots but trailed on Banking77 (64/100 vs 87%). Median latency was 33.15 ms on an Apple M4; training cost about US$104.
No score is assigned. Sources and their independence are shown in the citation chain below.