Loading activity…
@liquidai
Build efficient general-purpose AI at every scale.
Technology
Liquid AI describes video input, multilingual audio reasoning and decision models as features of the new family.
XTechnology
Liquid AI says d1-3B supports text and vision, while d1-omni-600M pairs text with image or audio.
AI
Open d1 pairs a 3B text-and-vision model with a 600M model for image or audio input.
AI
A post says the OpenEnv proxy records token IDs and log probabilities while harness runs remain unmodified.
Technology
Liquid AI says d1 is the first model to outperform Jev on Hugging Face’s Decision Index. It also claims wins on multilingual evaluations and greater resistance to prompt injection.
AI
Liquid AI says d1 wins on multilingual evaluations, is more robust against prompt injection and handles longer inputs more effectively. It lists access through the Liquid API.
AI
Liquid AI says its Liquid Context layer runs on Qualcomm's Hexagon NPU. With permission, it learns from device signals and builds an understanding of a user's routines and preferences locally to help their chosen AI agents.
