Reducto launches r-1, a one-pass document parser priced at 1 cent per page

Reducto has released r-1, a document parser that reads a full page in a single pass. Parsing here means turning a document into structured, machine-readable data. The company says r-1 replaces the multi-stage system it used before, runs faster, and costs less. It is available in preview through Reducto’s hosted Parse API, and the company has not released the model’s weights for others to run.

Reducto’s older Parse product ran three steps in sequence: optical character recognition, meaning reading text off an image; layout detection; and post-processing. Teams could add vision-language model passes on top, and each extra model call added delay. According to Reducto, r-1 folds text, tables, figures, layout, reading order, formatting, and grounding into one full-page pass. Every block of content comes back with a bounding box giving its position on the page.

The company frames the consolidation as the main point. Reducto says teams working on financial statements, insurance claims, or contracts often send files through several providers and bolt on their own cleanup code to reach usable accuracy, and that r-1 is aimed at that coordination cost rather than character accuracy alone.

Reducto reports a 20 percent drop in error rate for the r-1 preview compared with its own older pipelines. It also says r-1 did better than widely used hyperscaler products and large language models on complex documents in internal testing, naming Amazon Textract and Azure Document Intelligence as the incumbent category. Both results come from Reducto’s own tests. No public dataset or scoring code was released with the announcement, so the figures cannot be checked independently. On price, Reducto says legacy agentic models cost 3 to 6 cents per page depending on the work; r-1 is a flat 1 cent per page with no feature multipliers or credit charges added on.

r-1 requires the V3 version of the Parse API and is turned on with a configuration setting. A request that leaves that setting out still runs on the older Parse, so existing pipelines do not change on their own. New pipelines built in Reducto’s Studio interface default to r-1. There is no local copy to self-host, though Reducto’s platform separately offers customer VPC, on-premises, and air-gapped installs, with SOC 2 Type II attestation and HIPAA processing on higher tiers, per its security policies. Agentic processing stays for workflows that need custom prompts or advanced chart extraction; those pages route through the older pipeline and its results are added onto the r-1 output, which adds latency.

Reducto has named two next steps without dates: r-1 mini, for speed- and cost-sensitive work, and automatic per-page routing that picks a model for each page. Companies moving off another parser can request up to $5,000 in credits to run a side-by-side comparison. Whether the 20 percent figure holds up outside Reducto’s own testing will show in those comparisons and in customer reports, since no independent benchmark is available yet.