Nace AI Unveils Drex Max: A Model with 27 Billion Parameters

Nace AI presented a technical report on the Drex model and released Drex Max with 27 billion parameters. According to the developer, the model processes requests with a median latency of 183 ms – for comparison, some competing solutions have a latency of 2–3.5 seconds. In the Decision Index 0.4 test, Drex Max scored 67.52 points.

The model uses a single forward pass to generate a response, rather than sequential autoregressive processing. It accepts multimodal input data: text, images, JSON and video. The context window size reaches 262 thousand tokens. During training, RLAF methods, supervised fine-tuning, and loss functions aimed at improving result consistency were applied.

The lineup also includes a lightweight version of Drex with 9 billion parameters. The models are available through GitHub and Hugging Face under the Nace AI Open RAIL-M license, which may restrict certain commercial use cases.

Nace AI was founded in 2024 in Palo Alto. In May 2026, the company raised $21.5 million in a seed round led by Walden Catalyst with participation from General Catalyst. Among the company's products are Drex for Search and Drex for Finance.

Performance metrics and test results are provided by Nace AI and are not an independent assessment.