ScaleDown task-specific constrained-decoding models arrive on OpenRouter
- Source
- x.com
- Date
A family of constrained-decoding models from @scaledown_team is live on OpenRouter, each one purpose-built for a single task: • Summarize: turns long text, images, and PDFs into short, consistent summaries for context compaction and transcript digests • Extract: describe the fields you want in plain English and get clean JSON back, with confidence scores and source spans • Classify-1: set your own labels and get a calibrated probability for each, via our Decisions API 20% off for launch: $0.04/M input tokens, output free. https://t.co/YcCLlt2d2m
Context
The useful feature of ScaleDown's Extract model is that extracted values stay tied to where they came from. In OpenRouter's October 13 launch post, a caller describes the desired fields in plain English and receives JSON along with confidence scores and source spans, which point to the parts of the input each value came from. Launch announcement.
That gives a reviewer something concrete to check when turning documents into structured records: a low-confidence value, or a span that does not support its field, can be examined before the record is used. The announcement does not establish extraction accuracy. The confidence score is the model's own report, and a well-formed JSON record does not show that its values are correct.
- context compaction — Summarizing an agent's earlier conversation to free room in the context window so a long session can keep going.
- calibration — How well a model's confidence matches reality — a calibrated model saying "90% sure" is right about 90% of the time.
Checking sign-in…
Loading comments…