Liquid AI announced the release of d1-ultra, its most advanced multimodal decision model designed to process text and images in a single forward pass. Priced at $0.042 per million input tokens on the Liquid API, the model outputs probabilities for bounded questions without generating text tokens, allowing developers to avoid paying for output generation.

Built on a larger open-weight backbone with Liquid-specific customizations, d1-ultra aims to replace traditional LLMs for tasks like moderation, triage, LLM-as-judge, and guardrails. On the independent Artificial Analysis Decision Index, Liquid reports that d1-ultra ranks first with a score of 65, outperforming competing models like TypeSafe’s Jev 1.14 across 112 evaluations.

The model is available immediately on the Liquid API, with integrations for Vercel AI Gateway and OpenRouter scheduled soon. Billing covers input tokens and image patches directly, providing a predictable cost structure for high-volume developer workloads.

Why it matters

  • Offers developers an economical alternative to traditional LLMs for routing, moderation, and classification tasks.

  • Zero output token billing drastically reduces operational costs for high-throughput enterprise API calls.

  • Demonstrates non-generative decision models surpassing traditional architectures on specialized benchmark leaderboards.

Source: marktechpost.com