Decision models that return calibrated probabilities
Liquid d1.2 and d1.2-ultra are omnimodal decision models that return calibrated probabilities, so low-confidence answers can go to human review. This version adds audio input and five tasks: reading values, finding text spans, locating objects, estimating numbers with a range, and multi-select. Per Liquid AI, d1.2-ultra scores 75 on the Artificial Analysis Decision Index, two points above the previous best. On the Liquid API now; OpenRouter and Vercel coming soon.
Liquid d1.2 and d1.2-ultra are Liquid AI’s newest decision models. They add audio, five new tasks and what Liquid calls industry-leading accuracy.
A decision model takes an input and a question and returns an answer plus a probability that it’s right. Liquid AI says that probability is calibrated, meaning it should match how often the model is actually correct. That lets you set a threshold, like routing anything under 90% to a human for review, instead of trusting every answer equally.
What’s new:
Audio input, so you can ask about a sound or a language in a clip, or ask a question by recording it
Reading a value from an input, like a receipt total
Finding every passage of a kind in text, for example personal information
Locating objects in images, as one box or every instance
Estimating a number with a range
Picking every option that applies
Liquid AI says d1.2-ultra scores 75 on the Artificial Analysis Decision Index v2.1, the highest on the index and two points ahead of the previous best decision model. It also reports calibration error of 0.8 points for d1.2-ultra and 1.2 for d1.2, across 13 benchmarks and 22,505 questions, ahead of the other models it tested. The new-task scores come from Liquid’s own held-out test sets, for example 98% on reading receipt values and 94% on locating objects with d1.2-ultra.
Good fit if you’re building document processing, content moderation or routing workflows where you need to know how confident the AI is, not just what it answered.
“Decision models that return calibrated probabilities”
Liquid d1.2 launched on Product Hunt on November 17th, 2026 and earned 85 upvotes and 1 comments, placing #17 on the daily leaderboard. Liquid d1.2 and d1.2-ultra are omnimodal decision models that return calibrated probabilities, so low-confidence answers can go to human review. This version adds audio input and five tasks: reading values, finding text spans, locating objects, estimating numbers with a range, and multi-select. Per Liquid AI, d1.2-ultra scores 75 on the Artificial Analysis Decision Index, two points above the previous best. On the Liquid API now; OpenRouter and Vercel coming soon.
On the analytics side, Liquid d1.2 competes within Artificial Intelligence — topics that collectively have 484k followers on Product Hunt. The dashboard above tracks how Liquid d1.2 performed against the three products that launched closest to it on the same day.
Who hunted Liquid d1.2?
Liquid d1.2 was hunted by Kumar Abhishek. A “hunter” on Product Hunt is the community member who submits a product to the platform — uploading the images, the link, and tagging the makers behind it. Hunters typically write the first comment explaining why a product is worth attention, and their followers are notified the moment they post. Around 79% of featured launches on Product Hunt are self-hunted by their makers, but a well-known hunter still acts as a signal of quality to the rest of the community. See the full all-time top hunters leaderboard to discover who is shaping the Product Hunt ecosystem.
For a complete overview of Liquid d1.2 including community comment highlights and product details, visit the product overview.
Liquid d1.2 and d1.2-ultra are Liquid AI’s newest decision models. They add audio, five new tasks and what Liquid calls industry-leading accuracy.
A decision model takes an input and a question and returns an answer plus a probability that it’s right. Liquid AI says that probability is calibrated, meaning it should match how often the model is actually correct. That lets you set a threshold, like routing anything under 90% to a human for review, instead of trusting every answer equally.
What’s new:
Audio input, so you can ask about a sound or a language in a clip, or ask a question by recording it
Reading a value from an input, like a receipt total
Finding every passage of a kind in text, for example personal information
Locating objects in images, as one box or every instance
Estimating a number with a range
Picking every option that applies
Liquid AI says d1.2-ultra scores 75 on the Artificial Analysis Decision Index v2.1, the highest on the index and two points ahead of the previous best decision model. It also reports calibration error of 0.8 points for d1.2-ultra and 1.2 for d1.2, across 13 benchmarks and 22,505 questions, ahead of the other models it tested. The new-task scores come from Liquid’s own held-out test sets, for example 98% on reading receipt values and 94% on locating objects with d1.2-ultra.
Good fit if you’re building document processing, content moderation or routing workflows where you need to know how confident the AI is, not just what it answered.
Learn more: Announcement · Docs · Playground