Hyperscience advances Hypercell to unlock enterprise data for the agentic AI era. Fall 2026 release transforms complex documents into trusted, AI-ready data while applying the right combination of specialized and frontier models to deliver greater accuracy, control and cost efficiency. Sep 30, 2026 Prev Next 1 of 43,686 Hyperscience, a market leader in enterprise AI infrastructure software focused on Intelligent Document Processing (IDP), announced the Fall 2026 release of Hypercell, expanding the platform's ability to transform complex enterprise documents into trusted data for AI-powered workflows, applications, and agents. As enterprises move generative and agentic AI into production, access to models is no longer the only constraint. AI systems need accurate, contextual, governed enterprise data to reason and act reliably and much of that information remains trapped inside documents, forms, correspondence, images, and other unstructured content. This creates a new challenge for enterprises. Processing this information at scale requires more than sending every task to a frontier model. Organizations must balance accuracy, cost, latency, privacy, security, and governance while determining when to use specialized models, frontier models, deterministic validation, or human review. Hyperscience calls this the multi-model enterprise, an architecture in which organizations apply the right intelligence to each task rather than relying on a single model for every workload. The Fall 2026 Hypercell release advances this architecture by combining specialized document intelligence, vision-language models, frontier AI, validation, and human oversight to turn complex enterprise content into trusted, AI-consumable data, while giving organizations control over the economics and governance of AI at scale. A recent Hyperscience Harris Poll illustrates why this matters. Forty-seven percent of organizations are reevaluating their agentic AI strategies because of cost pressures, while 86% say data-quality issues are hurting agentic AI performance. At the same time, 89% are already using or developing workload routing that directs tasks to different models based on their characteristics. "Enterprise AI doesn't have a model problem. It has a data problem. The world's most powerful models can't reason over information they can't access, can't trust, or can't afford to process at scale. And some of the most valuable proprietary data in the enterprise is still trapped in documents," said Andrew Joiner, CEO, Hyperscience. "The Hyperscience Hypercell changes that. We turn complex document estates into trusted enterprise intelligence and apply the right combination of specialized AI, frontier models, validation and human judgment to each task. That's how enterprises move from impressive AI demonstrations to reliable AI economics and real-world outcomes." The Fall release advances Hypercell across three requirements for production enterprise AI: unlock more information, make that information trustworthy, and deliver it with enterprise economics and control. The innovations built into Hypercell deliver on these document intelligence requirements by enabling organizations to understand more, protect information, verify what AI extracts, expand access to information, and create high-quality data faster. Together, these capabilities help organizations move from simply processing documents to transforming the information within them into trusted inputs for AI-powered decisions and actions. Understand more: Enterprise information is increasingly complex, spanning structured forms, semi-structured documents, tables and unstructured content. ORCA 2, the next evolution of the Hyperscience Vision Language Model framework, expands Hypercell's ability to extract intelligence from this content with zero-shot table extraction and increased accuracy across structured, semi-structured and unstructured documents. ORCA 2 can also generate ground-truth data for downstream specialized and frontier models, helping organizations create higher-quality data while reducing the effort required to build and maintain AI models. Protect information: For enterprise AI to act on sensitive information, organizations need to control what data can be exposed downstream. New Redaction and Masking capabilities provide a purpose-built interface for reviewing and redacting PII across entire submission packages, including recurring entities such as names, Social Security numbers, dependents and dates of birth. New searchable rendered PDFs also ensure that sensitive information is removed from the underlying document content, while creating an indexable output that can support downstream search and Retrieval-Augmented Generation (RAG). Verify what AI extracts: Extraction alone does not make enterprise data trustworthy. Organizations need to establish whether information is accurate enough to support a downstream action. Address Verification & Validation Intelligence (AVVI) provides source-independent address parsing and validation, transforming complex address strings into structured components and validating them against Google Address Validation, enterprise reference data or third-party lookup services. This enables organizations to separate the extraction of information from the verification of that information without being locked into a single validation source. Expand access to information: Enterprise documents frequently contain valuable information in languages that may not be understood by the people responsible for processing them. Hypercell Document Translations brings in-line translation directly into document workflows, enabling organizations to process customer and constituent responses in 26 languages without introducing a separate translation workflow or expensive infrastructure. Translated results are immediately available during supervision and quality assurance, helping knowledge workers route documents to the appropriate workflow regardless of language. Create high-quality data faster: As organizations expand AI-powered workflows, the ability to create authoritative training and processing data becomes increasingly important. Auto-Labeling and Flexible Extraction combine ORCA segmentation models and frontier models to pre-annotate documents and assist with extraction, allowing keyers to populate fields by clicking or lassoing document segments. The result is a hybrid supervision experience that reduces manual labeling effort, accelerates deployment of new document workflows, and creates high-quality data for model training and continuous improvement. Together, these innovations make Hypercell more than a system for extracting information from documents. They provide an intelligence layer that helps enterprises understand, protect, verify, expand and continually improve the information flowing from their document estate into AI-powered workflows, applications and agents - while maintaining the human oversight and governance required for production AI. "Enterprise AI is moving beyond the assumption that one model can handle every task. Organizations increasingly need the flexibility to combine different models based on the requirements of the workload, while maintaining the accuracy, governance, and economics needed for production," said Alan Pelz-Sharpe, Founder, Deep Analysis. "This latest release reflects that evolution by giving enterprises new ways to improve document intelligence, accelerate the creation of high-quality data, protect sensitive information, and bring more complex workflows into production without adding unnecessary operational complexity."

Hyperscience has announced the Fall 2026 release of Hypercell, its platform that transforms complex enterprise documents into AI-ready data. The update addresses a key challenge as companies deploy agentic AI: much critical enterprise information remains trapped in unstructured documents. The platform combines specialized document intelligence, vision-language models, frontier AI, validation and human oversight. Hyperscience advocates a "multi-model enterprise" architecture, applying different AI models to specific tasks rather than using one model for everything. A recent Hyperscience Harris Poll found 47% of organisations are reevaluating agentic AI strategies due to cost pressures, whilst 86% report data-quality issues harm AI performance. Meanwhile, 89% are using or developing systems that route tasks to different models based on task characteristics. The release aims to balance accuracy, cost, latency and governance whilst processing documents at scale.