Nace AI has introduced Drex Max, a multimodal decision model designed for computer-use agents. Unlike generative models, Drex Max does not produce tokens. Instead, it processes a provided state and specific questions, returning a probability distribution for answers in a single HTTP call. The model supports three question types: yes/no, choice, and ordered scale.
The model is accessed via the POST /v1/systemone endpoint using the identifier drex-max. It accepts images as workspace files or base64 strings, which are referenced as plain text in queries. According to Nace AI, Drex Max achieves a median decision time of 174 ms per screenshot, while maintaining compatibility with all screenshot sizes.
Nace AI reports a median latency of 183 ms across its benchmark table. The company also cites performance figures of 34 ms per decision and sub-second latency at 262K tokens, though these were measured under different conditions. The model is trained on official splits and evaluated on held-out data.
Pricing for Drex Max is set at $0.06 per 1 million input tokens, with no charges for output tokens. New users receive a $25 signup credit, and failed requests are not billed. The model currently holds a score of 67.59 in Nace AI's internal evaluations.
