OpenRouter announced on 10 November 2026 that its Extractions API is live in alpha. Instead of free-form chat output, a request names the fields it wants from a document (text, image or PDF) and gets one typed value per field back.
There are nine field types: text, number, date, time, datetime, list, duration, estimate and forecast. Depending on the model, answers also carry a character span showing where the value was read, a confidence, and p_null, the probability that the document holds nothing for the field. A value becomes null above 0.5 by default, and the threshold can be set per field.
Three extraction models are available at launch, from Cloudflare (Clairic), Inception and ScaleDown (Extract). OpenRouter says the API is still in alpha, so details may change.
For customers: this is a new endpoint on the existing OpenRouter API. Documents sent to it go to the third-party model provider, so check that provider's data policy before sending sensitive files.