According to CoinMeta, the AI laboratory Supersonic Labs has released a lightweight decision-making model julia-1 with only 144.3 million parameters. This model does not generate long text; instead, it receives context, questions, and 2 to 20 candidate answers, and directly performs classification, scoring, and yes/no judgments. It can be used to select tools, make routing decisions, or determine the next steps for agent. Developed based on the multi-language encoding model mmbert-small, the model weighs about 550MB and is designed to have low hardware requirements. It has been tested on Apple M4, Intel, i5-1235U, and Samsung Android platforms. Officials state that the total cost of cloud GPU training and experimentation for julia-1 is approximately $104. The model weights, running code, complete evaluations, and source records are all available publicly, and it uses the Apache 2.0 license.