Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- Liquid AI releases d1-ultra, a multimodal decision model on its API at $0.042 per million input tokens
Liquid AI released d1-ultra, a decision model that reads text, images or both with one or more questions in a single forward pass and returns calibrated probabilities for yes/no, choice and score questions without generating tokens. Built on a larger open-weight base than d1 with Liquid-specific customizations, it targets harder tasks such as spotting a circuit-board defect against its spec. Liquid AI says it beats d1 and Jev 1.14 across 112 evaluations with better calibration, and outperforms other decision models on the Artificial Analysis Decision Index by at least 8 points. It is billed on input tokens only, with Vercel and OpenRouter availability coming soon.
- Liquid AI releases open d1-3B and d1-omni-600M decision models for edge devices
In a Hugging Face blog post, Liquid AI releases two open models in its d1 decision model family: d1-3B, built on LFM2.5-VL-3B and accepting text and images, and the experimental d1-omni-600M, which takes text with images or audio. Liquid AI says d1-3B scores 48.57 on Decision Index 0.2.1, the best under 10B, and answers a question in 16 ms on an NVIDIA Jetson AGX Thor and 50 ms on a Jetson Orin Nano.
- Liquid AI's d1 adds vision, matches GPT-6.1 Sol on four of six real applications
Liquid AI introduced d1, which it calls the most capable decision model and the first to extend decision capabilities to images. A decision model reads questions in one forward pass and returns probabilities without generating any tokens, with Noul, Choice, and Score question types. On six real applications, d1 matched or beat GPT-6.1 Sol on four and costs 19x to 200x less than the chat models tested. A text decision takes 200 to 300 ms, and on the public VisA dataset d1 sorted good and defective parts with 85-97% accuracy despite never being trained for the task.