Drex Max scores 67.6 on the public Decision Index and returns a visual decision in 183 ms
Synopsis
Nace released Drex Max, a decision model that reads screenshots, screencasts, scanned documents, charts or photos and, given a set of typed questions, returns a probability for every supplied option in one forward pass. A computer-use agent can face about thirty such small decisions in a single task, and each one previously meant sending a screenshot to a large frontier model and waiting seconds; Drex Max scores 67.6 on the public Decision Index 0.4.0 against 61.3 for Drex v2.0, with a median API latency of 183 ms and $0.06 per 1,000 decisions.
Interpretation
Decision models now take visual input directly: Drex Max accepts a screenshot, screencast, scanned document, chart or photo together with a set of typed questions and returns a probability for each candidate option in one forward pass. The prior route sent a screenshot to a large frontier model and waited seconds for it to write an answer; Drex Max targets the frequent small decisions inside an agent loop. On the public Decision Index 0.4.0, Drex Max scores 67.6, Drex v2.0 scores 61.3, Drex v1.5 scores 58.1, Jev 1.14 scores 60.7 and GPT-6 Luna Decisions scores 47.8.
It writes no text, running one pass over the input and scoring every option, so the cost of a call does not grow with the number of options. A generative model pays for every token it writes, so more options mean more output; a Drex Max decision costs one pass over the input. Median API latency is 183 ms at $0.06 per 1,000 decisions, against roughly 2.0 s and $3.33 per 1,000 for GPT-6 Sol and roughly 3.5 s and $10.17 per 1,000 for Claude Opus 5.5.
Perspective
The report covers the Drex Max release and benchmark tables published by Nace: decision, knowledge, reasoning, agentic and efficiency metrics, plus the perception, documents-and-charts, agentic and reasoning-and-safety groups of the vision decision index. The intended readers are researchers and engineers building computer-use agents; the material shows recordings of booking a flight and watching a live camera, without quantitative results for those demonstrations.
The material calls Drex Max the first decision model to lead the public Decision Index 0.4.0, but that index is a public leaderboard and not every system in the table is ranked on it. The vision decision index compares only against GPT-6 Luna Decisions, and Drex Max trails it on KIE and Hateful Memes. Latency figures are median API times including network, with approximate values marked by ~.