PatentEyeball
A prior-art search tool for patent attorneys that uses vision-language models to scan diagrams in patent filings and surface ones that look visually similar to a target invention, with explicit 'might be wrong' flags on uncertain matches.
Patent attorneys and IP paralegals doing prior-art searches
- Visual similarity search across USPTO, EPO, and WIPO patent diagrams using embedding models
- Atomic-perception checks: counts components, reads labels in figures, identifies orientations — the exact weak points PerceptionBench flags
- Side-by-side figure viewer with a 'confidence this diagram is relevant' badge per result
- Exportable prior-art report draft the attorney can edit before filing
PerceptionBench isolates the perception capabilities (counting, fine-grained recognition, OCR, depth) that patent image search most relies on — and shows frontier models are unreliable at exactly those tasks.
Prior Art Search AI market reached USD 1.62B in 2024 with multiple 2025 ranking articles devoted to AI prior-art tools and claims of 80%+ time savings, indicating strong practitioner pull.Prior Art Search AI Market Research Report 2033 ↗Top 7 Prior Art Search Tools for Patent Attorneys in 2025 ↗
Crowded: IPRally Image Search, XLScout Para-Picx, USPTO DesignVision (free, July 2025), Patsnap Eureka, Solve Intelligence, Orbit Intelligence all already offer image/diagram-based prior-art search.Introducing Image Search | IPRally Blog ↗USPTO DesignVision AI Image Search Tool ↗Visual Patent Search: Para-Picx for Image-based FTO Analysis ↗
Patent attorneys demonstrably pay for IP research platforms (Patsnap, Orbit Intelligence, Questel) and the AI patent-search market is USD 1.48–1.62B, but USPTO DesignVision is free to examiners and pressures pricing.Prior Art Search AI Market Research Report 2033 ↗Patsnap Pricing Plans ↗
Prior-art search demand is durable because patents keep filing, and uncertainty-flagging is a defensible UX hook, but frontier VLM perception is improving fast — the gap PatentEyeball exploits may close within a few model cycles.AI Patent Search Market Size | CAGR 12.5 Forecast 2033 ↗
Hard: PerceptionBench shows frontier VLMs are unreliable at counting, fine-grained recognition and OCR — exactly the perception tasks diagram prior-art search depends on, so a trustworthy product is difficult to ship without heavy human-in-the-loop.Prior Art Search: 7 AI Tools Ranked for Patent Professionals ↗