Liquid AI's d1.2-ultra tops Decision Index, adds audio input
TL;DR
- Liquid AI's d1.2-ultra scores 75 on the Artificial Analysis Decision Index v2.1, two points ahead of the previous best, per the company's own benchmarks.
- Across 13 benchmarks and 22,505 questions, Liquid reports calibration error of 0.8 for d1.2-ultra versus 2.7 for Jev 1.14, 3.7 for Perplexity's Decider v1.3, and 7.2 for GPT-6 Luna.
- The release adds audio input and five new API tasks, priced at $0.02 per million input tokens for d1.2 and $0.042 for d1.2-ultra.
Liquid AI released two decision models on November 16, d1.2 and d1.2-ultra, with the company claiming on its blog that d1.2-ultra "scores 75, the highest on the index and two points ahead of the previous best decision model" on the Artificial Analysis Decision Index v2.1.
The release adds audio input and five new API tasks. "d1.2 and d1.2-ultra also take audio as input," the post says, with the audio serving either as content to analyse or as the spoken question itself. The five new tasks cover reading values from inputs, finding passages in text, locating objects in pictures with bounding boxes, estimating a number with a range, and picking every option that applies. Liquid reports 98% accuracy on reading a value from a receipt for d1.2-ultra and 96% for d1.2, with both models at 94% on locating an object described in a photo.
On calibration, the post writes that "d1.2-ultra's calibration error is 0.8 points, compared with 1.2 for d1.2, 2.7 for Jev 1.14, 3.7 for Perplexity's Decider v1.3, and 7.2 for GPT-6 Luna" across 13 benchmarks and 22,505 questions. Pricing is set at $0.02 per million input tokens for d1.2 and $0.042 for d1.2-ultra. All of the scores and calibration figures come from Liquid's own reporting; no third-party replication is attached.
Shared on Bluesky by 2 AI experts
-
LiquidAI's d1.2 and d1.2-ultra (open-weight in the future, maybe) SOTA omnimodal decision models. d1.2 and d1.2-ultra both can process text, images, and audio with new output capabilities. www.liquid.ai/blog/d1-2
View on Bluesky →
Originally reported by liquid.ai
Read the original article →Original headline: d1.2: State-of-the-Art Omnimodal Decision Models | Blog