Nace.AI: Drex v2.0

Nace.AI: Drex v2.0 is a decision model from Nace.AI, and the larger successor to Drex v1.5. It is built as a 35B mixture-of-experts model with 3B active parameters, and it sits in the decisions category rather than the general chat category. If you are evaluating it, the important framing is that it is not a text generator in the usual sense: it is designed to take a described situation and return answers to specific questions about it.

The model accepts text input, including JSON, and produces decisions as output. The provider describes the intended pattern as giving the model a state (text or JSON) plus typed questions such as yes/no or multiple choice, and receiving the corresponding answers. The context window is 131072 tokens, and the maximum output is 117964 tokens. Because the output modality is decisions rather than free-form prose, you should expect structured answers to the questions you pose, not an essay or a chat reply.

That shape points to a fairly specific set of uses. Anywhere you have a state you can serialize as text or JSON and a set of discrete questions about it, Drex v2.0 is a candidate: triage and routing decisions, classification into a fixed set of options, yes/no gating checks, or multiple choice selection over a defined answer space. The large context window means the state you pass in can be long, so the questions can be grounded in a substantial body of input rather than a short snippet. It is less obviously suited to tasks where you want generated prose, summarization, or open-ended conversation, since those are not what its output modality covers.

You call it through api14's OpenAI-compatible API using the model id "nace-ai/drex-v2.0". Because the API is OpenAI-compatible, you can point an existing OpenAI client at api14 and send requests in the usual chat completions shape, with the model field set to that id. Note that the model lists no supported parameters, so you should not assume tuning knobs such as temperature or top_p are accepted; send the input state and your typed questions and rely on the defaults. The live price table on this page covers what a request costs.

On limits, the facts here are narrow. The model is text in and decisions out, so if your task needs image or audio input, or needs generated text as the output, another model will fit better. Likewise, since no supported parameters are listed, if your pipeline depends on adjusting sampling behavior, check that before committing. Within those bounds, Drex v2.0 is aimed squarely at structured decision tasks over long text or JSON states.