Onton’s Ontology 1: A Neurosymbolic AI Model Redefining E-commerce Search Accuracy

Revolutionizing Product Discovery with Advanced AI

The central development is this: In the rapidly evolving landscape of online retail, finding the perfect product often feels like searching for a needle in a haystack. Traditional e-commerce search engines, while powerful, frequently struggle with the nuances of human language and complex, conversational queries. Enter Onton, a San Francisco-based search and discovery company, which has recently unveiled Ontology 1. This groundbreaking neurosymbolic search model promises to transform how we interact with online catalogs, boasting an impressive accuracy rate that significantly surpasses industry giants like Google Shopping and Amazon.

Meanwhile, Ontology 1 isn’t just another incremental update; it represents a fundamental shift in search technology. By combining the pattern recognition capabilities of neural networks with the logical reasoning of symbolic AI, it delivers unparalleled precision for complex, multimodal product searches.

Unpacking the Benchmark: Ontology 1’s Dominance

To demonstrate its capabilities, Onton subjected Ontology 1 to the rigorous Subtext-Decor-90 benchmark. This test involved 90 unique text queries, with the top 10 results from Onton, Google Shopping, and Amazon being evaluated by three independent LLM judges (Claude Opus 4.8, Gemini 3.1 Pro, and GPT-5.5). The results were striking:

• Ontology 1: Mean precision@10 of 0.630
• Google Shopping: Mean precision@10 of 0.543
• Amazon: Mean precision@10 of 0.469

In practical terms, This translates to Ontology 1 being approximately 2.7 times more accurate than the world’s best e-commerce search engines, a remarkable feat considering it indexed only about 1% of the catalog size of its competitors. Ontology 1 outright won 52 of the 90 queries, while Google secured 19 and Amazon 16.

The Power of Neurosymbolic Reasoning

Traditional e-commerce search often relies on keywords and predefined attributes like size, price, or material. This approach falls short when users express nuanced needs, such as searching for a “pet-friendly sectional” or “furniture that fits a strangely deep windowsill.” These queries don’t map neatly to existing filters, leading to frustrating search experiences.

For example, Ontology 1 takes a radically different approach. Instead of merely matching keywords, it builds an explicit, inspectable “world model.” For a “pet-friendly sectional,” it doesn’t blindly trust a seller’s label.

Instead, it reasons from objective properties like fiber type, weave, and construction, flagging claims that contradict product data. It also considers the source’s credibility, acknowledging that some listings might try to game algorithms or reviews could be inauthentic.

When faced with an unfamiliar concept like “pet-friendly,” Ontology 1 decomposes it into verifiable properties, such as cleanability and durability, then identifies indicators like polyester upholstery. This learning is then continuously reused and refined for subsequent, related queries, creating a self-improving search loop.

Deployment, Availability, and Target Audience

That said, Ontology 1 is not a model you can simply download. It’s live and accessible to end-users on Onton.com. For businesses looking to integrate this advanced search capability, Onton is offering partner access on a case-by-case basis, particularly for teams developing on the “agentic web.” There is currently no public API, pricing tier, or open-source checkpoint for the model itself, indicating a strategic focus on collaborative partnerships rather than broad public release.

Who Benefits Most?

Ontology 1 is particularly suited for:

• Mid-market and enterprise retailers: Especially those whose existing search solutions struggle with lengthy, complex queries.
• Marketplaces: Seeking to enhance product discovery and relevance.
• Agentic-commerce platforms: Building advanced shopping agents that require robust grounding layers.

Interestingly, While currently focused on the home decor and furniture industries (the only vertical Onton indexes at present), the underlying methodology is designed to generalize beyond e-commerce, suggesting future applications in non-product data search with minimal reconfiguration.

Key Applications Include:

• Conversational and multimodal site search
• Moodboard-driven discovery experiences
• Negation-heavy filtering (e.g., “show me sofas *without* leather”)
• Listing and review trust scoring
• Grounding layers for advanced shopping agents

The Infrastructure Behind the Intelligence

At the core of Ontology 1’s knowledge graph is Ograph, a custom-built graph database. Onton reports impressive performance metrics for Ograph, with a single core outperforming SuiteSparse:GraphBLAS running on 14 cores—a roughly 100x throughput per core improvement. Furthermore, a GPU build of Ograph runs 43x faster than its CPU counterpart, with early optimizations hinting at a potential 1000x speedup.

Where Ontology 1 Still Seeks Growth

However, Despite its impressive capabilities, Ontology 1 isn’t infallible. Failure cases tend to cluster around highly functional-spec queries where Amazon’s extensive category metadata currently dominates.

For example, queries like “lamp that won’t wake my partner if I read at 3 am” or “something to put on a weirdly deep windowsill” saw Amazon outperform Ontology 1. Onton attributes this to its currently smaller, single-vertical, non-sponsored index and anticipates that its continuous self-learning loop will steadily narrow this gap over time.

Expert Perspective

From an industry angle, the clearest signal around Neurosymbolic Search is how it may influence ontology. The story reads less like a one-day spike and more like a marker of broader movement.

The next phase will depend on how quickly teams, regulators, or customers react. In practice, that gives Neurosymbolic Search room to reshape expectations across quot over the near term.

For readers focused on practical impact, the best next step is to watch what changes around search once attention turns into execution.

Frequently Asked Questions

Why does Neurosymbolic Search matter right now?

Revolutionizing Product Discovery with Advanced AIThe central development is this: In the rapidly evolving landscape of online retail, finding the perfect product often feels like searching for a needle in a haystack.

What broader change could Neurosymbolic Search signal?

Traditional e-commerce search engines, while powerful, frequently struggle with the nuances of human language and complex, conversational queries.

What should the market watch next around Neurosymbolic Search?

Enter Onton, a San Francisco-based search and discovery company, which has recently unveiled Ontology 1.

Key Takeaways

• Ontology 1 achieved a precision@10 of 0.630 on the Subtext-Decor-90 benchmark, significantly outperforming Google Shopping (0.543) and Amazon (0.469).
• It won 52 out of 90 queries while indexing a fraction (approximately 1%) of its competitors’ catalogs.
• The model’s neurosymbolic architecture utilizes an inspectable knowledge graph to break down vague queries into verifiable properties.
• While judge reliability was modest (Krippendorff’s alpha 0.465), all three judges consistently ranked the engines in the same order.
• Availability is product-first, live on Onton.com, with partner access granted on a case-by-case basis. There are no open weights or public API for direct integration.

Source: https://www.marktechpost.com/2026/08/02/onton-releases-ontology-1-a-neurosymbolic-search-model/