Nace AI Open-Sources Drex Max v1, a 27B Multimodal Decision Model for Local Inference
The startup released the weights of Drex Max v1, a calibrated decision model that returns option probabilities in a single forward pass, with GGUF builds that run on Apple Silicon.
Nace AI opened the weights of Drex Max v1 on November 1, making its decision model available for local inference. Unlike generative chat models, Drex Max takes a state — text, JSON or images — plus a typed question and returns calibrated probabilities for each option in a single forward pass, without generating free-form text. The dense 27B model uses a hybrid architecture with linear and full attention layers and supports a 262,000-token context.
The release includes bf16 weights, Q8_0 and Q4 GGUF builds, a llama.cpp fork and Ollama support; the quantized version runs on Apple Silicon with at least 24 GB of memory. The model is listed at 64 on the third-party Artificial Analysis Decision Index. Licensing is under Nace.AI Open RAIL-M, which carries use restrictions that enterprises need to review before commercial deployment.