Nace AI has released Drex Max v1 as an open-weights 27-parameter decision model designed to run locally on consumer hardware, including Apple Silicon Macs. The release includes bf16 weights, a Q8_0 GGUF, an affine Q4 checkpoint, and forks of llama.cpp and Ollama. This allows developers to handle routing, classification, and policy checks entirely on local machines.
Drex Max processes typed questions about a given context, returning probabilities for up to 255 options in a single forward pass without generating text. The architecture features a dense 27B backbone with 64 layers, incorporating a native vision tower and a 262,144-token window. On performance benchmarks, Nace AI reports that the model scores 67.52 on the public Decision Index 0.4, outperforming its predecessor Drex v2.0.
Local deployment is supported across multiple runtimes, with the affine Q4 checkpoint targeting Macs with 24GB or more of RAM, and the Q8_0 GGUF requiring 48GB. Testing on Apple Silicon hardware showed minimal performance loss compared to NVIDIA H100 execution, enabling offline deployment for enterprise routing tasks.
This release matters because it lowers the barrier for developers to run advanced, context-aware decision models on local hardware without relying on cloud APIs.