Veriwire
← Back to the wire

Meet Drex Max: Nace AI’s Multimodal Decision Model for Computer-Use Agents with a Score 67.59

AnnouncementModelOct 20, 2026

Nace AI has released Drex Max, a multimodal decision model that reads text and images and returns calibrated probabilities without generating tokens. Nace reports 67.59 on the public Decision Index 0.3.2, 6.9 points above TypeSafe's Jev 1.14, plus 92.2 on JevBench versus 86.7. It trails Jev 1.14 on MMLU-Pro (72.5 vs 82.1). Available via hosted API only; parameters undisclosed; scores are self-reported.

Why it matters

Nace AI has launched Drex Max, a multimodal decision model aimed at agents that operate computers and browsers. Unlike generative models, it produces no text; it accepts a state plus typed questions of three kinds, including yes/no and label-choice formats, and returns probability distributions in a single call. Nace reports it leading the public Decision Index and JevBench against TypeSafe's Jev 1.14, though Jev 1.14 remains ahead on MMLU-Pro, HLE, and ARC-Challenge. The model is available only through a hosted API, with no open weights, and its parameter count is undisclosed. It accepts up to eight images per request in common web formats, with each image consuming tokens in every question. Nace also reports latency figures for screenshot decisions that are faster than Jev 1.14's, though the article notes different setups behind separate speed claims. All performance numbers are self-reported by Nace, and cross-vendor comparisons are not controlled, so readers should treat rankings cautiously.

Receipt № 24831 source · awaiting confirmation ◐

Evidence

1source· awaiting independent confirmation

No score is assigned. Sources and their independence are shown in the citation chain below.

Citation chain · 1 source

01med
www.marktechpost.comBloghttps://www.marktechpost.com/2026/10/19/nace-ai-drex-max-multimodal-decision-model/This source supplies the technical and operational details of the Drex Max release, including its typed question interface, image handling limits, latency claims, and self-reported benchmark comparisons against Jev 1.14.
PRIMARY
TypeSafeCompanyNace AICompanyDrex MaxModelJev 1.14Model

Related

Oct 19GenRecs Tokenization: Transforming Ads Recommendation at Meta with Stable Model VocabularyOct 19MiniMax Just Launched M3.1 Flash: 428B MoE Agentic Model at $0.15/M InputOct 19Mistral AI Releases Timestral: A 209M Forecasting Model Family That Outperforms Google’s TimesFM-3
Canonical: https://www.marktechpost.com/2026/10/19/nace-ai-drex-max-multimodal-decision-model/