LiquidAI has released two new decision models, d1-3B and d1-omni-600M, which return typed answers in a single forward pass with zero output tokens by reading directly from the model's distribution over options.

  • d1-3B is a 3B parameter multimodal model built on LFM2.5-VL-3B that achieves 48.57 on the Decision Index 0.2.1, outperforming larger models like Decider 35B-A3B.
  • d1-omni-600M is a 587M parameter decision model built on LFM2.5-Encoder-350M that supports vision-language and audio-language inputs in a single pass.
  • Both models operate without text generation or parsing, with d1-3B completing decisions in 8 ms on an NVIDIA RTX 4090.

These releases provide fast, edge-sized decision-making capabilities for multimodal tasks.