Use Nace AI Drex Max API for Multimodal Computer Agent Guidance

Nace AI has released Drex Max, a new multimodal decision model designed to help computer-use agents understand text and screenshots. The model returns calibrated probabilities for specific questions rather than generating free-form text. According to MarkTechPost, Nace reports a score of 67.59 on the public Decision Index 0.3.2. This performance places it 6.9 points above TypeSafe’s Jev 1.14 model. The release targets builders who need efficient, typed responses for agent interactions.

In plain English

A decision model differs from standard language models because it does not generate open-ended text. Instead, you provide the model with a specific state and named questions. Each question must be one of three types: yes or no, a choice from provided labels, or a score on an ordered scale. The model returns a probability distribution for each answer in a single HTTP call. It can only answer using the options you explicitly give it. This approach ensures structured outputs for automated systems.

View the original post on X

The background

Drex Max is the most accurate model in Nace’s Drex family and the only one that reads images. It uses the same POST endpoint as other Drex models, selected by specifying the model ID. Images are included in an array, with every question seeing the same images before the state and questions. Users refer to images as Image 1 or Image 2 in plain text. Each request allows up to eight images, with a total size limit of 32 MiB.

What changes now

The model is available through a hosted API only, with no open weights or quantized builds announced. It supports up to 262,144 tokens per question, including image tokens. Each image costs up to 1,602 tokens in every question. Performance benchmarks show Drex Max scoring 98.2 on BFCL tool calling and 92.2 on JevBench. However, it scored lower on HLE and MMLU-Pro compared to Jev 1.14. All scores are self-reported by Nace AI.

Key terms explained

System administrators should understand that decision models provide structured outputs rather than natural language. Teams usually integrate these models to automate specific tasks where predictable responses are critical. Check whether your agent workflows require image analysis or simple text queries. Evaluate if the cost per token aligns with your operational budget. Ensure your API infrastructure can handle the image payload limits. Verify that the model’s response format matches your existing data structures.

What to do and how to stay safe: Nace AI

• Review your agent architecture to determine if a decision model replaces current text-generation workflows.
• Test the hosted API with sample images to verify latency and token costs fit your requirements.
• Monitor the decision index scores for updates that may affect model selection strategies.
• Ensure your security policies allow external API calls for image processing and data transmission.

Step-by-step guide: Stop AI Chatbots Leaking Sensitive Business Data

General security guidance from the Vector Update 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 AI reports a cost of $0.06 per 1 million input tokens for the Drex Max model.

Does Drex Max support local deployment?

No, Drex Max is available via hosted API only, with no open weights or local builds announced.

How does Drex Max compare to Jev 1.14?

Drex Max scored 67.59 on Decision Index 0.3.2, which is 6.9 points higher than Jev 1.14’s score of 60.7.