About Liquid d1.2
Liquid d1.2 and d1.2-ultra are decision models that return an answer alongside a calibrated probability of confidence, helping teams distinguish straightforward cases from those that need human review. The models support audio input and tasks such as reading values from receipts, locating objects in images, finding text spans, estimating numbers with ranges, and selecting all applicable options. This makes them useful for workflows where businesses need both an extraction or classification result and a way to set a review threshold. For example, a team could extract totals from receipts, flag uncertain results for manual checking, locate products in image assets, or route potentially sensitive text for moderation; the models are available through the Liquid API, with OpenRouter and Vercel listed as upcoming integrations.
What you can use it for
- Extract values such as totals from receipts and documents
- Route low-confidence AI decisions to human reviewers
- Locate objects in images for catalog and asset workflows
- Find sensitive information or other specified spans in text
- Classify audio and image inputs for business workflows
- Estimate quantities with ranges and select multiple applicable labels
