Nace AI releases Drex Max multimodal decision model for computer-use agents
Nace AI has launched Drex Max, a multimodal decision model tailored for agents that navigate computers and browsers. Unlike text-generating models, Drex Max accepts text and up to 8 images per request, returning calibrated probability distributions for structured questions rather than generating text tokens. It supports context of up to 262,144 tokens per question through a hosted API, with no…
Key points
- Drex Max processes text and images to return calibrated probabilities for agent choices without generating text.
- Nace AI self-reports a score of 67.59 on Decision Index 0.3.2, outperforming TypeSafe's Jev 1.14 by 6.9 points.
- The API costs $0.06 per 1M input tokens with free output, supporting up to 262,144 tokens per question.
Nace AI reports that Drex Max scored 67.59 on the public Decision Index 0.3.2, which the company states is 6.9 points higher than TypeSafe's Jev 1.14. It also achieved 98.2 on BFCL tool calling and 74.5 on the Decision Index Vision public suite, while trailing Jev 1.14 on benchmarks such as MMLU-Pro (72.5 versus 82.1) and HLE (19.0 versus 19.4). All benchmark figures are self-reported by Nace AI.
The model is available through the Nace Console under the pinned identifier drex-max. Pricing is set at $0.06 per 1 million input tokens, with output tokens unbilled, and Nace AI claims a median latency of 174 milliseconds per screenshot decision.
Model page: Jev 1.14 →
The story so far
5 episodes →- Nace AI releases Drex Max multimodal decision model for computer-use agentsthis story
Meet Drex Max: Nace AI’s Multimodal Decision Model for Computer-Use Agents with a Score 67.59
MarkTechPost · 20 October 2026
Loading the full article…
This text was published by MarkTechPost and written by Michal Sutter. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Continuity Research releases Tess-4.5-27B open-weights coding model · 1 src
- Tel Aviv-based doubleAI releases debut code execution model Maxwell and raises $200M-$220M · 3 src
- OpenAI says its model is better at COBOL, Fortran, and Visual Basic · 1 src
- Fal launches FLUX 3 Max video model, 22× faster than Seedance 2.5 · 1 src
- Google makes Gemini 4 Argon generally available · 2 src
Comments
via GitHub Discussions