The development

What happened

Liquid AI released d1-3B and the experimental d1-omni-600M on Hugging Face. The models answer structured questions in one forward pass: d1-3B accepts text and images, while d1-omni-600M accepts text with images or audio.

Key facts

  • Both open-weight models became available on 7 October; d1-omni-600M is an early research release.

  • The models produce structured decisions in a single forward pass rather than generating text token by token.

  • Liquid AI reports d1-3B single-question latency of 16 ms on Jetson AGX Thor, 26 ms on AGX Orin and 50 ms on Orin Nano.

  • The announcement provides no vision or audio benchmark results and no speed measurements for d1-omni-600M.

The context

Why it matters

For developers evaluating local classifiers or routing decisions, this offers a different approach from generating and parsing a text response. The reported latency is specific to Liquid AI's tests; longer inputs and images take longer. Multimodal capability alone does not establish accuracy for a particular task. Open weights also require checking the model license before deployment.