Nace AI introduces drex max, a high-performance multimodal decision model for computer agents

Nace AI has launched Drex Max, a multimodal decision model designed for computer-use agents. The model processes text and images, providing calibrated probabilities without generating text. It achieved a score of 67.59 on the Decision Index 0.3.2, surpassing TypeSafe’s Jev 1.14 by 6.9 points. Drex Max is the most accurate model in the Drex family and the only one capable of reading images. It operates via the POST /v1/systemone endpoint with the model identifier 'drex-max'. The model handles three question types: yes/no, choice, and score. It returns probability distributions in a single HTTP call, based on user-provided options. Nace claims Drex Max processes images efficiently, with a median latency of 174 ms per screenshot, compared to Jev 1.14’s 253 ms. Drex Max also accepts all screenshot sizes, unlike Jev 1.14, which rejects 15.9% due to size constraints. Pricing is $0.06 per 1M input tokens, with no output token charges. New users receive $25 in credit, and failed requests are free.