Liquid AI launched d1-ultra on October 15, 2026, its most capable multimodal decision model to date. Unlike large language models that generate text, d1-ultra answers questions with calibrated probabilities in a single forward pass and produces no output tokens.
What happened?
The model accepts text, images, or both as input and answers three types of questions: yes/no (noul), choice among labels, or a score on a scale. According to the company, it was built for harder real-world cases, such as spotting defects on circuit boards against specs or checking insurance damage claims against customer photos.
Liquid AI states that d1-ultra leads all eight categories in its internal 112-evaluation suite and scores 65 on the Artificial Analysis Decision Index, beating Jev 1.14 (TypeSafe) and other decision models by at least 8 points. The company also highlights the lowest calibration error among models tested, allowing reliable thresholds for automation (high confidence) versus human review (low confidence).
Why it matters
Decision models are useful for classification, routing, moderation, and triage tasks where generating long text is unnecessary and expensive. By not charging for output tokens and processing everything in one step, d1-ultra offers a more efficient and predictable alternative for agent workflows and production systems. Multimodality expands use cases to combined visual and text analysis.
Availability
The model is available today on the Liquid AI API at $0.042 per million input tokens (no output cost). Images are counted as tokens (1.5 tokens per 32×32-pixel patch). Support on Vercel AI Gateway and OpenRouter is coming soon. No open weights have been announced; access is API-only. The playground is at d1.liquid.ai.
Liquid AI had previously released smaller d1 family versions and the LFM2.5-24B-A2B for on-device agents. d1-ultra represents the jump to the top of the decision-model category.
Sources
Transparency: This content was created, edited, or reviewed with the assistance of artificial intelligence. Information was cross-checked with public posts on X and sources available on the internet. Consult the original sources to verify the full context.
By GeekikiBot