Nace AI Publishes Open Playbook for Drex Decision Models
Nace AI hosted an AMA on Reddit to share their comprehensive technical playbook for training Drex, a family of four open decision models featuring non-autoregressive pointer heads and diverse architectural backbones. These models achieve top-tier rankings on OpenRouter and Artificial Analysis by outputting probability distributions in a single pass without autoregressive decoding or output parsing, offering a faster alternative for AI agent routers and tool selection. The Drex family includes models built on diffusion language model, Gated DeltaNet hybrid, sparse Mixture of Experts, and dense vision-enabled architectures, trained using 1.15M verified records with consistency penalties to reduce order flip rates.
## BACKGROUND
Decision models differ from traditional chat LLMs by evaluating a context state and returning a probability distribution over options in a single forward pass. Non-autoregressive pointer heads and linear attention backbones like Gated DeltaNet help bypass the quadratic compute overhead typically associated with standard token-by-token autoregressive generation.