- Published on
A Neurosymbolic Search Model That is 2.7x More Accurate than the World’s Best E-commerce Search Engines
- Authors
- Name
- DeepHorus
San Francisco-based search and discovery company Onton has released Ontology 1, a neurosymbolic model for complex, conversational, and multimodal product search. In a 90-query benchmark, it achieved a mean precision@10 of 0.630, beating Google Shopping’s 0.543 and Amazon’s 0.469. It did this while indexing roughly 1% of their catalogs.
Conventional e-commerce search assumes intent maps onto categories and attributes like size, price, material, and brand. Onton argues this interface has barely changed in nearly 30 years. Ontology 1 reasons from more objective properties such as fiber, weave, and construction, and flags claims the product data contradicts.
The model builds an explicit, inspectable world model rather than absorbing patterns into weights. When it has no account of “pet-friendly,” it works out the answer through cleanability and durability, then polyester upholstery as an indicator. Onton released the benchmark Subtext-Decor-90 with code and data. Three LLM judges—Claude Opus 4.8, Gemini 3.1 Pro, and GPT-5.5—scored top-10 results. Onton won 52 queries outright, Google 19, and Amazon 16.
Krippendorff’s alpha across judges was 0.465, so absolute scores are noisy, but all three placed engines in the same order. Ontology returned fewer than 10 results on three queries; excluding those empty slots raised its score to 0.665. The model is live at Onton.com, but there is no public API or open checkpoint; partner access is case by case.
Target users are mid-market and enterprise retailers whose relevance stacks lose on long, requirements-heavy queries. Small catalogs see less benefit because the failure mode scales with catalog size and listing noise. The methodology generalizes beyond e-commerce, Onton says.
Failure cases cluster on functional-spec queries where Amazon’s category metadata dominates. For instance, “lamp that won’t wake my partner if I read at 3am” scored Onton 0.4 vs Amazon 0.9. Onton attributes this to catalog breadth and its single-vertical, non-sponsored index.
Onton currently indexes only home decor and furniture, but plans to expand. The model aims to serve the agentic web, providing grounding layers for shopping agents. Adoption will likely resemble partnerships rather than a pip install. As conversational and multimodal search grows, Ontology 1 could reshape product discovery.
