A supplier contract with no standard layout can now be extracted without labelled training data. ABBYY's new PHOENIX Plus does it with generative models the company selects, tunes and hosts inside its Vantage platform.
PHOENIX Plus applies zero-shot extraction to unseen, long-tail and highly variable documents. Prompts define the output schema in place of training data.
ABBYY supplies pre-engineered prompts for common cases. For a contract with varied layout, terminology and clauses, that removes much of the work of designing and testing prompts from scratch.
The models go beyond extraction. They can enrich and reason over validated data, interpret free-form text, identify inconsistencies and generate recommendations to support downstream decisions.
PHOENIX Plus is the generative half of ABBYY's PHOENIX portfolio. PHOENIX Core provides the deterministic foundation: optical character recognition, layout analysis, classification, extraction and document-specific intelligence.
ABBYY says the hybrid approach uses deterministic Document AI where precision and consistency are essential. Generative AI is applied where interpretation and reasoning add the most value.
The generative layer runs alongside existing Vantage controls. These include data validation rules, confidence thresholds, human-in-the-loop review and source traceability.
PHOENIX Plus is hosted by ABBYY, with data kept in the same region as the customer's Vantage data. Customers add it to their Vantage subscription at a fixed annual cost.
That means no separate model vendor contract, no infrastructure to provision and no token-based billing to manage. Customers can still connect OpenAI, Google Gemini, Anthropic Claude, Mistral or custom models through Vantage's bring-your-own-model option.
ABBYY cites a logistics customer that used PHOENIX Plus to read handwritten and printed text from complex forms. It achieved 99 per cent accuracy on the first pass.
"Enterprises today want the benefits of generative AI in their document workflows but require the compliance and auditability that enterprise-grade technology provides," said Bruce Orcutt, Chief Product and Marketing Officer at ABBYY.
"Providing a model that works inside your existing Vantage pipeline and managed entirely by ABBYY, PHOENIX Plus is the GenAI layer that enterprises want with the predictability and control that production automation demands," Orcutt said.
ABBYY first previewed PHOENIX, a zero-shot extraction approach built on small language models, at its Ascend event in 2024. The company launched FineParser, a CPU-only document parser, in September.
ABBYY told IT Brief that PHOENIX Plus would initially be available through an API. The company also plans to bring its DocLang format to Vantage.
ABBYY cited a fund administration company that processes around one million financial documents with complex tables each year. It was running ABBYY's technology within about a month, according to Director of Product Marketing Slavena Hristova.
PHOENIX Plus is available now to select ABBYY Vantage customers. ABBYY says its account teams will assess whether individual use cases qualify.