Liquid AI Releases d1-ultra: A Multimodal Decision Model That Scores 65 on the AA Decision Index

Liquid AI has launched d1-ultra, a multimodal model designed specifically for decision-making tasks. Unlike general-purpose language models, d1-ultra accepts both text and image inputs and returns calibrated probability scores in just one forward pass, eliminating the need for multiple inference rounds.

The model achieved a score of 65 on the Artificial Analysis Decision Index, a benchmark that evaluates how reliably a model can produce well-calibrated probabilistic judgments. The pricing is set at $0.042 per million input tokens, positioning it as a cost-efficient option for applications that require structured decision outputs.

The single-forward-pass design is notable because it reduces latency and computational overhead compared to approaches that require iterative sampling or multi-step reasoning. This makes d1-ultra potentially suitable for real-time decision pipelines in areas such as content moderation, risk assessment, or automated triage systems.

Liquid AI has been building models optimized for efficiency and edge deployment. d1-ultra extends that philosophy into the decision-making domain, where calibrated confidence scores are often more valuable than raw text generation. Whether the model's benchmark performance translates to strong real-world reliability will depend on broader third-party evaluations.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.