Nace.AI: drex-max

Nace.AI: Drex Max is the flagship multimodal decision model in the Nace.AI product family. Built as a 27B dense model based on Qwen3.8-27B, it uses a non-causal full attention mechanism and 255-direction readout heads. Instead of generating free text, it answers typed questions in a single forward pass: yes/no questions return the probability of yes, choice questions with 1 to 255 options return the most likely option with per-option probabilities and confidence, and ordered-level questions with 1 to 255 levels return the expected level with per-level probabilities, a legend and confidence. It never returns answers outside the options provided. The model accepts text, JSON and images, and each question can reference all images in the request. It is compatible with TypeSafe System One and shares the same request and response format as Jev. Long states above 8,192 tokens are truncated to the most relevant parts, and only tokens actually read are billed. Drex Max is designed for decision-making in computer and browser use agents, reading information from screens, charts, forms, photos and video, customer support urgency classification, routing inquiries to departments, and assessing customer frustration levels. It is licensed under the Nace AI Open RAIL-M License (modified) and is available on OpenRouter. Fully automated decision-making is prohibited in employment, credit, housing, insurance, healthcare, education, immigration and law enforcement without meaningful human review, as are discrimination based on legally protected characteristics, medical, legal or financial advice and decisions without qualified human oversight, and use for military, weapons, mass surveillance, fraud or spam purposes.