Nace.AI has open-sourced Drex 1.5, a 9B decision model built for agents and backend workflows. Unlike conventional language models, Drex 1.5 does not generate text. It reads a state and a set of typed questions, then returns a probability for every option. Nace reports a score of 58.08 on the public Decision Index 0.3.1βthe top result under 10B parameters. Weights are available on Hugging Face, and a hosted version is live on OpenRouter.
TL;DR
- Size: 8.95B parameters (dense), bf16 weights about 18 GB. Context is 16,384 tokens by default, up to 131,072.
- Runs on: 1 CUDA GPU in bf16 (tested on a 24 GB A10G). A Q8_0 GGUF (about 9.5 GB) runs on Apple silicon and CPU.
- Performance: 58.08 on Decision Index 0.3.1 (public), within the board's tie band of Jev 1.13.0 (57.96).
- Best for: Routing, selection, and scoring tasks where an agent must choose among discrete options rather than generate prose.
Why a Decision Model Matters in 2026
As agentic AI systems mature in 2026, the bottleneck is shifting from text generation to reliable decision-making. Multi-agent orchestration, tool selection, and workflow routing all require fast, calibrated choices among a finite set of actions. Drex 1.5 is purpose-built for that niche: it takes structured state and typed queries, then outputs probabilities over candidate options, sidestepping the latency and unpredictability of free-form generation.
Availability
Weights are published on Hugging Face under the Nace AI organization. A hosted endpoint is available on OpenRouter for teams that prefer not to self-host. The Q8_0 GGUF variant enables deployment on consumer hardware, including Apple silicon and standard CPUs.
This article is based on the original announcement and specification from Nace AI.
via MarkTechPost