Reducto Releases r-1 Document Parsing Model Cutting Errors 20%

The digital document processing landscape shifts as Reducto officially introduces r-1, an advanced parsing model designed to streamline how complex files are handled. The core innovation replaces multi-stage legacy pipelines with a unified, single-pass processing architecture.

Legacy parsing workflows typically run OCR, layout detection, and post-processing as separate stages, adding latency with every additional model call. In contrast, r-1 integrates text, tables, figures, layout, reading order, formatting, and grounding into a single full-page pass, returning bounding boxes that tie every block back to its exact page position.

Consolidating document parsing into a single pass addresses a major orchestration burden for development teams. Instead of stitching together multiple third-party providers and adding costly post-processing layers, r-1 handles complex tables and irregular layouts natively while preserving spatial accuracy, making it vital for regulated sectors like finance and legal.

Engineering teams working with dense financial statements, insurance claims, or contracts often rely on multiple disparate services to achieve usable accuracy. r-1 specifically targets this orchestration cost rather than focusing solely on raw character accuracy metrics.

Stock photo for illustration only, not from the actual event

Reducto reports a 20% reduction in error rates for the early r-1 preview against its own legacy agentic pipelines. Internal evaluations also indicate that r-1 outperformed commonly used hyperscaler products and large language models on complex documents, benchmarking against incumbent baselines like Amazon Textract and Azure Document Intelligence.

"Legacy Parse runs OCR, layout detection, and post processing as separate stages, with optional agentic vision language passes layered on top."

Reducto Documentation

On pricing, legacy agentic models previously ranged between 3 and 6 cents per page depending on workload complexity. r-1 is priced at 1 cent per page all-in, removing feature multipliers or credit costs previously required to achieve high accuracy tiers.

The r-1 model is currently deployable in preview through Reducto's hosted Parse API on V3 via a configuration flag, with no open weights or local checkpoints available for self-hosting. Furthermore, Reducto has signaled upcoming releases including r-1 mini and automatic per-page model routing.

Source: MarkTechPost