$0.06/M Tokens: Drex Max Hits 67.59 Decision Index, Tops Jev for Multimodal Browser Agents
The new multimodal model offers calibrated probabilities for browser agents at a cost of six cents per million input tokens.
Key points
- Drex Max achieves 67.59 on Decision Index 0.3.2, surpassing TypeSafe’s Jev 1.14.
- The model processes text and up to eight images per request via a hosted API.
- Pricing is set at $0.06 per million input tokens, excluding image token costs.
Nace AI released Drex Max, a multimodal decision model designed for agents that interact with computers and browsers. According to MarkTechPost, the model reads text and screenshots to return calibrated probabilities rather than generating open-ended text.
The system scores 67.59 on the public Decision Index 0.3.2. This performance places it 6.9 points above TypeSafe’s Jev 1.14. Nace reports that Drex Max is the most accurate model in its Drex family and the only one capable of reading images.
View the original post on X
Technical specifications and benchmarks
Drex Max operates exclusively through a hosted API on the Nace Console. No open weights or quantized builds have been announced. The model accepts up to 262,144 tokens per question. It processes images by resizing them to a maximum of 1,638,400 pixels. Each image costs up to 1,602 tokens in every question.
Performance metrics show strengths in tool calling and agent benchmarks. Drex Max scores 98.2 on BFCL tool calling and 92.2 on JevBench. However, it scores lower on HLE at 19.0 and MMLU-Pro at 72.5. These figures are self-reported by Nace on its Drex page.
Impact on agent builders
The model is intended for builders creating computer-use agents. It allows for one cheap, typed call per click. Users send a state and named questions to the model. Each question must be one of three types: noul, choice, or score. The model returns a probability distribution for each answer in a single HTTP call.
Organizations using this technology can integrate visual context into their decision-making workflows. The API supports up to eight images per request. Images can be uploaded workspace files or inline base64 entries. This capability enables agents to analyze screenshots alongside textual data.
Defender's checklist
Organizations should verify that their agent workflows handle image data securely. Review API usage limits and token costs for image-heavy tasks. Ensure that external models only receive necessary visual context. Monitor for unexpected behavior in agentic tasks.
Operational impact
Teams usually need to assess whether new AI components introduce data leakage risks. Check if the model processes sensitive screenshots or internal documents. Verify that third-party APIs comply with your security policies. Ensure that staff understand the model’s limitations in handling complex visual data. Coordinate with legal and compliance teams to review vendor contracts.
What to do and how to stay safe: Nace AI
- Review API documentation for data handling and privacy policies.
- Implement strict access controls for systems using external AI models.
- Monitor agent outputs for hallucinations or incorrect decisions.
- Plan for potential integration changes as models evolve.
Step-by-step guide: Implementing AI Usage Policies at Work: A Step-by-Step Framework
General security guidance from the InfoSec Ledger newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is the cost of using Drex Max?
Nace reports a cost of $0.06 per million input tokens for Drex Max.
Can Drex Max generate text?
No, Drex Max returns calibrated probabilities for predefined questions and does not generate text.
How many images can Drex Max process per request?
The model supports up to eight images per request, with a total size limit of 32 MiB.
