Nace AI open-sources Drex-Max v1: 27B multimodal decision model scoring 255 options in one pass
techNmak · x · 2026-10-30
Nace AI open-sourced Drex-Max v1, a 27B multimodal "decision model" for AI agents. Instead of generating tokens, it takes a state (text, JSON, images) plus questions with candidate answers and returns probabilities for every option in a single forward pass — up to 255 options, no generated text. Use cases include inspecting a screenshot to pick which UI element to act on, or deciding if a tool result needs another step.
The architecture is a dense 27B with 64 layers combining 48 Gated DeltaNet layers. It runs 100% locally on Apple Silicon (M4/M5) and via cloud inference; the family also includes Drex v2.0 (Qwen3.6-35B-A3B backbone), distilled v1.5 and Drex DLM, all runnable with llama.cpp/Ollama/MLX under a Nace.AI Open RAIL-M license.
More from Models
- EachLabs launches Lockstep recast model claiming 75% cost cut and better quality — miilesus · 2026-10-30
- ChatGPT cited a 300+ page government legal PDF it retrieved on its own — lilyraynyc · 2026-10-30
- Best MLX model for French-to-English translation on an M2 Mac? — greenreddits · 2026-10-30
- AI-driven cyber ops near the sophistication peak set by GTG-1002's Claude Code campaign — Jsevillamol · 2026-10-30
- Cognition FrontierCode chart pits Mistral Large 4 vs Chinese open models, sparking debate — onetwoval · 2026-10-30
- NVIDIA's Nemotron 3.5 Super hits 51.78% on SWE-bench Pro with 4.6x fewer active params — dair_ai · 2026-10-30