Document-processing AI Reducto launches r-1 mini: $1 document parsing per 1,000 pages
Reducto announced the launch of r-1 mini in a post on X on November 5, 2026, through Adit (@aditabrm). The company said the model is the first product to offer truly premium parsing at $1 per 1,000 pages while not carrying a premium price, outperforming options that cost up to 25 times as much.
Adit pointed out that models are becoming smarter, faster, and cheaper, but the OCR industry has not kept up. r-1 mini is positioned as premium parsing capability offered at a low price, suitable for large-scale document processing. The company previously launched r-1 in September 2026, at about $10 per 1,000 pages, and r-1 mini now lowers the price by another order of magnitude.
The company’s official blog said that in its internal evaluation, r-1 mini, at a price of $1 per 1,000 pages, outperformed some full-featured alternatives that cost 3.75 times as much and had accuracy 16 percentage points lower, and it led other parsers by more than 7 percentage points at prices below $10. The evaluation was based on the company’s in-house ParseArena benchmark, and the results are self-reported by the company.
Original source:
Title: Models are becoming smarter, faster, and cheaper, but the OCR industry hasn’t kept up. Publisher: Adit (@aditabrm), Reducto Publication date: 2026-11-05 URL: https://x.com/aditabrm/status/2118387155974414710 Source nature: firsthand post Verification status: Full post and associated replies were read for this report, verification date 2026-11-05; the accompanying blog was cross-checked Supporting content: Supports all core facts in the body, including the product name, pricing, performance claims, and publisher
ABAB AI Insight
Adit is a document-processing-related person at Reducto, and the company focuses on turning complex documents into AI-usable structured data. In September 2026, Reducto launched r-1, saying it was more accurate and faster than the strongest previous agentic OCR model, with costs as low as one-sixth of that model, and priced at 1 cent per page, or $10 per 1,000 pages. This r-1 mini lowers the price to $1 per 1,000 pages, making it a cost-optimized version under the same technical approach.
On the capital and business path, Reducto is based on a self-developed unified architecture: it first builds a high-precision model, then trains a smaller model to reduce inference costs. The flow of funds and resources is reflected in bringing the capabilities previously used for high-end agentic parsing down to large-scale, low-priority document scenarios, so enterprise customers do not have to skip large volumes of documents because of cost. The company blog mentioned that r-1 has helped customers solve edge cases and reduce costs, and that r-1 mini further opens up archives and pipelines that previously could not be fully processed because the per-unit price was too high.
Comparable cases include traditional OCR service providers such as Amazon Textract and Azure Document Intelligence, as well as emerging parsing tools such as LlamaParse and Extend. The similarity is that all try to trade off accuracy and cost; the difference is that Reducto uses a single end-to-end model to cover text, tables, and layout at the same time, and actively pushes prices down to the level of traditional raw OCR rather than maintaining a high-end premium. The company’s self-reported evaluation shows that r-1 mini achieved a higher win rate at a lower price in its benchmark, but no large-scale independent third-party validation has been publicly disclosed.
This release changes the cost threshold for document parsing, not the intelligence ceiling of AI models themselves. $1 per 1,000 pages makes full archive processing, research agents, enterprise knowledge retrieval, and other scenarios that were previously limited by unit price economically viable; for Reducto, it means exchanging price elasticity for greater processing volume. The observation metric is whether customers move from selective parsing to full-scale parsing in actual usage, and whether independent benchmark verification of its accuracy claims emerges.
ABAB News · Laws of cognition
The price of document parsing has dropped from $10 per 1,000 pages to $1, bringing low-priority archives that were previously skipped into the range of processable items.
Self-reported benchmarks show that a low-priced model can outperform higher-priced alternatives, but before independent verification, this cannot be equated with proven industry leadership.
Under a unified architecture, first building high precision and then lowering costs is the path for document processing to shift from a premium tool to infrastructure at scale.