Liquid AI added a new model called d1-3B to Hugging Face. The model answers named, typed questions over a state that can be text, JSON or images, and it does so in one forward pass without generating tokens. Users can run it with llama.cpp and query it via a /v1/systemone HTTP endpoint. Example calls show how it handles a billing‑refund request and how it identifies cats in a photo.
Why it matters
Developers can integrate fast, token‑free question answering into support bots or image‑based workflows without needing large language model generation.
