Liquid AI has introduced d1, a new decision model designed for structured choices instead of text generation. It provides calibrated probabilities across fixed outcomes in a single call, with no output tokens generated.
The model supports three primitives: Noul (yes/no with probability), Choice (selects one option with confidence), and Score (rates input on a rubric). These primitives cover most bounded decision tasks.
d1 is available now through the Liquid API under the name d1:free. It is an API-only model, with no options for self-hosting or training. This makes deployment simple and predictable.
Compared to large language models, d1 offers predictable latency, no schema errors, and usable uncertainty. It is ideal for tasks like moderation, ticket routing, and ranking.
Using d1 can lower costs by avoiding output token billing and reduce response times. It helps teams make consistent, threshold-based decisions at scale.
Why it matters
This model enables faster, cheaper, and more reliable decision-making in AI workflows.
What to do
Evaluate if your tasks are bounded and fit the primitives supported by d1. Use the API for deployment and testing.