Liquid AI Releases d1.2 and d1.2-ultra: Omnimodal Decision Models That Top the AA Decision Index
Summary
- Liquid AI released d1.2 and d1.2-ultra, omnimodal decision models that return calibrated probabilities over fixed answer sets with zero output tokens.
- The d1.2-ultra model scored 75 on the Artificial Analysis Decision Index v2.1, securing the top position among tested decision models.
- The models accept text, images, and audio inputs up to 65,536 tokens per question across 128 questions per API call.
- Five new answer types were added: value reading, span extraction, object localization, numerical range estimation, and multi-label choice selection.
- Weights and parameter counts are not yet publicly available, with models accessible exclusively through Liquid AI's hosted API.
Why it matters
Decision models offer an efficient alternative to traditional generative LLMs by outputting precise probabilities instead of text tokens. This reduces generation costs and improves calibration for automated routing and classification tasks.