Jev is overtaken for the first time, and Liquid AI releases its decision-making model d1.

Insight Beating AI Flash News: Frontier edge AI R&D firm Liquid AI has launched its first self-developed decision-making model, d1. Like Jev, d1 does not generate long texts; instead, it directly performs selection, judgment, and scoring, enabling agents to select tools, handle routing, and conduct audits. Liquid re-ran evaluations using Hugging Face’s decision model comprehensive benchmark Decision Index 0.2.1. d1 scored an overall 58.9, outperforming Jev’s 57.9. Across five major categories, d1 leads in art judgment, language understanding, and retrieval classification; it trails Jev slightly in tool calling, and lags by 8 points in knowledge reasoning. Liquid notes that d1 is the first model to surpass Jev in this evaluation suite. d1 is compatible with Jev’s three decision-making modes: Choice (selecting one from multiple options), Noul (true/false judgment), and Score (scoring against standards). A single request can complete multiple judgments at once, returning probabilities and structured results directly—no intermediate text generation and parsing required. d1 is now available via Liquid’s API, with a free tier named `d1:free`. The model’s weights and parameter size have not been disclosed, and Liquid plans to integrate it into OpenRouter.