Drex Max v1 reads text, JSON or images and returns a probability for each option instead of writing text. Weights are now public, and a text-only build runs on a Mac with 24 GB or more of memory.
Most AI answers by writing. Drex Max instead scores every option you give it in one pass, like a judge holding up score cards. Nace AI released the 27B model’s weights on Oct 30 for local use, including on Macs.
Nace reports 67.52 on its public Decision Index, above its own Drex v2.0 (61.25). The license is free for individuals, researchers and organizations under $1 million in revenue or funding.
Key Capabilities:
- Runs locally: A 4-bit build runs on 24 GB Macs, and the Q8 version needs 48 GB.
- Text-only locally: The GGUF, Q4 and MLX local versions read text only, not images.
- Open RAIL-M license: Not standard open source, and larger companies need a deal with Nace.