Liquid AI launches d1-ultra decision model via API

Liquid AI released the multimodal decision-making model d1-ultra on October 15, 2026, and made it available through its API, according to aitimes.com. The model accepts text and images, or both together, and produces a result in a single forward pass without generating separate output tokens.

Alongside its decision, d1-ultra reports probabilities for each option. According to Liquid AI, the model outperformed TypeSafe AI's Jev 1.14 across all eight of the 112 evaluation areas it ran. On Artificial Analysis's Decision Index, d1-ultra scored 65 points and ranked first.

Pricing is $0.042 per million input tokens, the same input price as Jev, with no separate charge for output tokens. Images are billed as input tokens at 1.5 tokens per 32x32 pixel patch, so a 1024x1024 image counts as 1536 tokens.

Liquid AI did not disclose the model's parameter count or the name of the open-weight model it is based on, and no downloadable weights or quantized versions were announced. The company said support via the Vercel AI Gateway and OpenRouter is planned.

Earlier context

Liquid AI released the previous model, d1, earlier in October 2026. On October 13, 2026, the first Decision Index evaluation was published, in which Jev 1.13 scored 58 points and the original d1 led on knowledge accuracy with 41 versus Jev 1.13's 35, but scored only 2 on avoiding wrong answers versus Jev's 35. In the same evaluation, GLiDE from Passino Labs and Microsoft's Decision-1 each scored 54 points.