Liquid AI Open-Sources d1 Family of Real-Time Decision Models
On October 8, Liquid AI released Open d1, open-sourcing two multimodal models from its d1 decision model family, with NVIDIA partnering on the rollout.
Confirmed
- d1-3B: text + vision, built on LFM2.5-VL-3B, delivers decisions in a single forward pass; per @RespectJaded15, it tops the sub-10B category on the Decision Index v0.2.1 benchmark.
- d1-omni-600M: supports text + image or text + audio, aimed at real-time decision-making.
- Extremely broad deployment range: runs on everything from NVIDIA DGX data centers and RTX workstations to edge devices like Jetson, with multiple posts citing on-device inference as low as the 10ms level.
- NVIDIA Robotics' official account reshared congratulations on the release.
Why it matters
- Traditional language models need multi-step reasoning to produce decisions, while the d1 family outputs decisions directly in a single forward pass—a good fit for real-time control scenarios like robotics.
- Open weights plus data-center-to-edge portability lower the barrier to deploying robot decision models, and reinforce the signal of Liquid AI and NVIDIA deepening their robotics ecosystem partnership.
2026-10-08 ~ 2026-10-08 · 7 related posts
- Liquid AI releases Open d1: open-weight 3B and 600M multimodal decision models — JosephJacks_ · 2026-10-08
- Most decisions need a fast model, not a big one: Liquid AI open-sources d1-3B — JosephJacks_ · 2026-10-08
5 near-duplicate retellings: JosephJacks_ · JosephJacks_ · JosephJacks_ · JosephJacks_ · JosephJacks_