Liquid AI Open d1 is a family of open-weight decision models for developers and edge engineers who need fast, structured answers on device instead of token-by-token generation. Released on October 7, 2026, the d1 models do not produce text at all; they score a fixed set of answer options in a single forward pass, which makes them a fit for real-time routing, moderation, and agent control loops. Both checkpoints are downloadable from Hugging Face with no usage restrictions.
Core Features
- d1-3B (3.12B params) takes text and images and returns an answer in one pass.
- d1-omni-600M (587M) adds audio for text-plus-image or text-plus-audio inputs.
- Latency of 8 ms on an NVIDIA RTX 4090, 16 ms on Jetson AGX Thor, and 50 ms on Orin Nano.
- Native llama.cpp support across Apple, AMD, Qualcomm, and NVIDIA hardware, including NVFP4.
- A Decision Index v0.2.1 score of 48.57, ahead of every model under 10B.
Use Cases
- Real-time content moderation running d1-3B over a live camera feed.
- On-device routing of support tickets to billing, technical, or fraud teams.
- Robotics and agent loops that need a quick structured choice without a GPU server.
Pricing
The models are open-weight and free to download, fine-tune, and deploy with no restrictions. Weights ship on Hugging Face alongside a System One Arcade demo space; Liquid AI does not sell a paid API tier, so cost is whatever your own edge hardware already runs. It benchmarks against Decider 35B-A3B at a fraction of the size.
Pros and Cons
- Pro: sub-10 ms decisions on consumer GPUs make on-device inference practical.
- Con: it answers a fixed set of choices you define, not open-ended prompts.
Our Take: Best for teams shipping low-latency multimodal decisions to the edge; trade-off is that you must specify the questions and options up front rather than prompting freely.
