LiquidAI Launches d1-3B and d1-omni-600M Decision Models for Edge Applications

On October 7, 2026, LiquidAI released two new decision models: d1-3B and d1-omni-600M. The d1-3B model achieved a score of 48.57 on the Decision Index 0.2.1, making it the best decision model under 10 billion parameters, according to LiquidAI. It supports both text and images, while the d1-omni-600M, which has only a quarter of the parameters of the Decider 2B, supports text and images or text and audio, scoring 78.4, surpassing Decider 2B's score of 77.1.

The d1-3B model operates at speeds of 16 ms on the NVIDIA Jetson AGX Thor, 26 ms on Jetson AGX Orin, and 50 ms on Jetson Orin Nano. In contrast, the d1-omni-600M is designed for lower resource environments while maintaining competitive performance. The d1-3B achieved a mean score of 82.9 across seven public datasets, the highest in its category, while the d1-omni-600M scored 78.4.

Performance metrics for d1-3B include:

• SQuAD 2.0 Score: 83.3
• Civil Comments Score: 93.3
• MASSIVE Intent Score: 86.9
• PubMedQA Score: 68.3
• BoolQ Score: 86.3
• XNLI Score: 85.6
• PAWS-X Score: 76.4

In terms of hardware performance, the d1-3B model runs at 8 ms for a single question on the NVIDIA RTX 4090, and 21 ms for three questions. The AMD MI325X shows similar performance, with 9 ms for one question and 14 ms for three questions. This launch follows previous advancements in accelerating vision-language models with LFM2.5-VL-DSpark and faster inference techniques with LFM2.5-DSpark.

LiquidAI's new models provide edge computing practitioners with robust decision-making capabilities, enhancing the performance of AI applications in resource-constrained environments.