LiquidAI: d1.1

LiquidAI: d1.1 is a decision model built on a larger open-weight base with Liquid-specific customizations. Rather than producing prose, it answers structured questions directly: yes/no questions return a probability between 0 and 1, choice questions return a probability for each label, and score questions return an expected position across ordered levels. It can take several questions about the same state in one request and computes all option probabilities in a single forward pass with zero output tokens. It accepts text and image input and produces text output. The model is aimed at fast structured decision-making inside software. Typical uses include classification, routing, scoring, support ticket triage, content moderation, response evaluation, image inspection, and gating agent steps. Text decisions take 200 to 300 milliseconds, which is fast enough for real-time applications, and the model is positioned as a replacement for costly calls to large language models when only a structured decision is needed. It reports equal or better performance than GPT-6.1 Sol and Claude Opus 5.5 in 4 of 6 real applications, while responding substantially faster than both on all tasks. Other reported strengths include strong multilingual evaluation results, high resistance to prompt injection, and effective handling of longer inputs. In interactive tests, adding the screen alongside text raised a Tetris score from 70 to 81; in Wordle it read the board directly from screenshots and solved 12 of 12 games in an average of 3.8 guesses; and in Quick, Draw! it recognized 5.2 of 6 doodles from 62 words.