Nace’s Drex Max v1 model card makes the 27-billion-parameter decision model available for local deployment. Instead of generating prose, it calculates probabilities for up to 255 choices given a context. Mac guidance specifies at least 24 GB of memory for an MLX-based four-bit route or 48 GB for an eight-bit GGUF route; local Mac execution is text-only, unlike the NVIDIA and Python paths with image support. The weights expand deployment options beyond the earlier API offering, but OpenRAIL-M restrictions apply, and the advertised maximum context does not guarantee practical operation on every Mac.