NaceAI launches Drex, a decision model outperforming Jev

Jev: Jev is an established decision model that outputs probabilities for given options without generating prose. Drex is positioned as a direct improvement over Jev, matching the same request format while advancing performance on decision benchmarks. It serves as the reference point for the public Decision Index leaderboard that evaluates models on closed-choice tasks. Drex: Drex is a compact decision model created by NaceAI that processes inputs to return probability distributions over user-provided options in a single pass. It is built as a small diffusion model with RLAF and is intended for applications requiring direct decisions rather than text generation or parsing. NaceAI presents it as a faster, more targeted alternative for agent stacks and related decision pipelines. NaceAI: NaceAI is the organization that develops specialized AI models focused on decision-making tasks. It recently launched Drex as a new model explicitly positioned for workloads involving routing, tool selection, and policy checks where options are predefined. The company emphasizes efficient, non-generative outputs for practical agent and system integration. Decision Models: Decision models deliver option probabilities directly in one pass, eliminating the need to generate and parse text that is common in generative approaches. Agent Efficiency: Specialized decision models can streamline agent routing, tool selection, and policy checks by focusing compute only on choosing from known options.