What changed
OpenRouter launched an Extractions API in alpha for pulling named, typed fields from text, images and PDFs. Responses include values, source spans and confidence scores, with launch models from Cloudflare, Inception and ScaleDown.
Why it matters
The endpoint provides a common schema-driven surface for document extraction across specialist models. Alpha status and model-dependent confidence mean it should augment, not replace, deterministic validation on financial or regulated documents.
What to do
Test a versioned schema on a labeled document set, verify spans and null behavior, and add deterministic checks plus human review for high-cost fields before production use.
Caveat
This is an alpha API with three launch models. Confidence scores are model outputs rather than guarantees, and the announcement does not establish service levels, cross-model consistency or independent accuracy.