Liquid AI’s d1-ultra Ranks First on a Decision-Model Benchmark

Liquid AI’s d1-ultra has taken the top position on a decision-model benchmark, putting a system that assigns probabilities to predefined answers first across tests of knowledge, reasoning, tools and safety.

The model accepts text, images or both and returns probabilities for answers supplied with a question. That differs from open-ended text generation: d1-ultra evaluates a set of choices and indicates how likely each answer is. Potential applications include detecting product defects in images and reviewing damage claims alongside customer photos.

The Decision Index combines four evaluation areas: knowledge and hallucination, reasoning, tool calling, and safety classification. Each receives 25% of the total weight. Within the knowledge-and-hallucination category, accuracy and non-hallucination rate each account for 12.5% of the overall index.

The ranking reflects performance on those benchmark tasks rather than every kind of language-model work. The evaluations are conducted independently, and the index’s result does not establish how the model will perform in every real-world application.

Liquid AI announced d1-ultra on Oct. 15. It is available through the Liquid API for $0.042 per 1 million input tokens, with billing based on input and no charge for output tokens.