This is basically a zero-shot classifier that can accept raw text (or structured text) as an input, and is able to classify that text as accurately (they claim) as a frontier-level LLM. I have workflows this would be useful for, looking forward to it showing up on OpenRouter.

exactly right!