Jev was surpassed for the first time, Liquid AI releases decision model d1.

Beating AI News Flash: Frontier edge AI R&D company Liquid AI has released its first self-developed decision model, d1. It does the same type of work as Jev, not generating long text, but directly making choices, judgments, and scoring, and can select tools, perform routing, and audit for Agents.

Liquid reran the evaluation according to Hugging Face's Decision Index 0.2.1, a comprehensive benchmark for decision models. d1 scored 58.9 overall, surpassing Jev 1.13's 57.9. Among the 5 major categories, d1 leads in artistic judgment, language understanding, and retrieval classification, is slightly lower than Jev in tool calling, and trails by 8 points in knowledge reasoning. Liquid claims that d1 is the first model to surpass Jev in this evaluation suite.

d1 is compatible with Jev's three decision modes: Choice selects one from multiple options, Noul judges true or false, and Score rates according to criteria. A single request can also complete multiple judgments at once, directly returning probabilities and structured results without the need to first generate text and then parse it.

Currently, d1 is available through the Liquid API, with the free version being `d1:free`. The model weights and parameter scale have not yet been disclosed, and Liquid also plans to integrate it into OpenRouter.