November 2026 Summaries

Reducto introduces Pairwise AI Ranking (PAIR), an open evaluation framework for document parsers that uses an LLM judge to compare two blinded parses against an original page image and determine which more faithfully preserves information. The company argues that conventional rule-based and ground-truth benchmarks can be brittle because they often penalize harmless differences in markup or formatting while missing consequential errors such as incorrect table relationships, values, reading order, or key-value pairings. PAIR compares each candidate parser with a shared baseline across benchmark pages, repeats judgments three times with shuffled output order, and calculates win rates with ties counted as half wins. In a third-party annotation study of 300 comparisons, the LLM judge agreed with final human decisions 84.0% of the time when both selected a winner and 91.6% when human reviewers were unanimous. Controlled edits also found that the judge consistently treated formatting-only changes as ties while detecting meaningful content changes, such as altered digits or swapped table cells and paragraphs, at high rates. Reducto has published the judge prompt and ParseJudge tool so organizations can evaluate parsers on their own documents without creating document-specific rules or extensive human-labeled datasets.

Reducto has launched r-1 mini, a document-parsing model priced at $1 per 1,000 pages that it says delivers higher accuracy than competing low-cost and some premium parsers costing up to 25 times more. Designed to make large unstructured document collections economically usable by AI applications, the model offers full-layout and agentic parsing capabilities for use cases such as enterprise search, web crawling, and research agents, while its larger r-1 model remains available for higher-accuracy, lower-latency needs involving complex charts, scans, diagrams, equations, and multilingual documents. Reducto evaluated parsers through ParseArena, a 2,000-page cross-industry benchmark assessed with its open-source PAIR pairwise AI judging framework, which compares outputs against original pages rather than relying on rigid format-based scoring. According to the company’s results, r-1 mini outperformed several offerings from Azure, Amazon Textract, Google Gemini, Mistral, LlamaParse, Datalab, and Extend, while producing fewer serious errors than some major cloud alternatives. Reducto is also making r-1 generally available, offering migration credits for teams changing providers, and expanding access to individual users through personal-email signups and free credits.

Reducto argues that despite rapid advances in AI models, adoption in industries such as healthcare, finance, and insurance remains constrained by entrenched, inefficient workflows and information locked in documents. The company focuses on extracting data from complex files such as PDFs, invoices, bank statements, and medical records, which traditional OCR-based document-processing systems often handle only through brittle, narrowly tailored pipelines and human review. Reducto says it has prioritized human-level accuracy for difficult cases including rotated pages, complex tables, and handwriting, but acknowledges that this focus has historically increased cost and latency. Its broader goal is to become an interface between human-created documents and AI models by offering accurate, accessible tools across the full document lifecycle, including parsing, reading, editing, generation, and production infrastructure. The company plans to introduce a new product intended to advance this expanded strategy.