Onton, a San Francisco-based search and discovery company, has unveiled Ontology 1, a groundbreaking neurosymbolic model poised to revolutionize complex, conversational, and multimodal product search. This innovative system achieved a remarkable mean precision@10 of 0.630 on a 90-query benchmark, significantly outperforming industry giants Google Shopping (0.543) and Amazon (0.469). Notably, Ontology 1 accomplished this while indexing only about 1% of its competitors' catalog sizes. While not available as a public API or open-source model, Onton is offering partner access for teams building on the agentic web, signaling a new era for intelligent discovery.
Unlike traditional e-commerce search engines that rely on rigid categories and attributes, Ontology 1 employs a sophisticated neurosymbolic approach. Instead of merely matching keywords, it constructs an explicit, inspectable "world model" to reason about product properties. For instance, for a query like 'pet-friendly sectional,' it infers meaning from objective characteristics such as fiber, weave, and construction, rather than blindly trusting seller labels. It also flags contradictory data and weighs source credibility. This allows it to deduce complex concepts, treating unknown terms as gaps to be filled through logical inference, with learned insights continuously reused for subsequent queries.
This release holds significant implications for developers and researchers. Ontology 1 demonstrates a powerful paradigm shift from pattern absorption to explicit reasoning, offering a path toward more accurate, explainable, and adaptable search systems. The accompanying release of the Subtext-Decor-90 benchmark, complete with code and data, provides a valuable resource for advancing research in neurosymbolic AI and complex information retrieval. It underscores the potential for AI to move beyond simple keyword matching, enabling truly intelligent and context-aware discovery experiences.