Patsnap Launches Industry-First Patent Bench for AI Patent Novelty Benchmarking

Patsnap Launches Industry-First Patent Bench for AI Patent Novelty Benchmarking

“Why IP professionals need AI benchmarking like Patent Bench”

New Era in Patent Intelligence: On September 10, 2025, Singapore-based innovation intelligence leader Patsnap unveiled a pioneering solution aimed at reshaping how artificial intelligence tools are evaluated in intellectual property (IP) work. The newly launched Patent Bench represents the industry's first quantitative benchmarking tool specifically designed for AI-assisted patent novelty searches. This strategic development marks a significant step toward establishing greater consistency, accuracy, and domain-specific reliability in the application of AI to the patent landscape.

What Is Patent Bench?

Patent Bench is not merely another benchmarking tool; it is a comprehensive and structured evaluation mechanism tailored to the complex needs of patent professionals, researchers, and R&D departments. Unlike generic AI benchmarks, it is purpose-built to assess AI tools within the context of intellectual property, taking into account the intricate legal and technical language that dominates patent documentation. According to Jeffrey Tiong, Founder and CEO of Patsnap, Patent Bench goes far beyond traditional text-generation assessments. “It drills down to deep, technical, and legal jargon often used in patent work,” he explained, emphasizing that the benchmark caters to the real-world challenges faced by IP professionals.

Benchmarked AI Models

At its launch, Patent Bench evaluates the performance of three AI tools:

• Patsnap’s Novelty Search AI Agent
• ChatGPT-o3 (enhanced with web search capabilities)
• DeepSeek-R1 (also equipped with web search)

These models were selected for their varied capabilities in processing patent data, offering users a comparative view of how different AI systems perform when tasked with identifying novel technical features and executing precise search strategies.

The Technology Behind the Benchmark

Patsnap’s Novelty Search AI Agent, one of the evaluated tools, stands out for its specialized development. Built on an open-source base model, the agent has been refined with patent-specific knowledge to handle the unique linguistic and logical challenges of patent searching. It leverages Retrieval-Augmented Generation (RAG) — a framework that blends real-time data retrieval with generative AI — to minimize hallucinations and enhance the relevance and reliability of search results.

This enables the AI agent to:

• Understand complex patent claims and technical descriptions
• Apply contextual reasoning within legal frameworks
• Consistently deliver accurate and high-utility search outputs

Addressing a Longstanding Industry Gap

One of the most notable aspects of PatentBench is that it fills a long-standing void in the industry. Tiong pointed out that many clients and stakeholders have expressed the need for a robust and credible benchmark tailored specifically to IP. General-purpose large language models, while impressive in broader contexts, often fall short in patent-specific applications.

“General LLM benchmarks are not robust enough to assist in patent work,” Tiong noted. “The task requires deep expertise across three domains: law, technology, and business. Oftentimes, they miss the nuances that professionals care about in claims, are unfamiliar with the diverse and evolving regulations across patent offices, and lack evaluation standards tailored to real-world needs like patent searches.”

Looking Ahead: Beyond Novelty Searches

While the current iteration of Patent Bench is focused on novelty search evaluation, its long-term vision is far more expansive. Patsnap plans to extend the benchmark to cover a broader range of patent-related tasks, including:

• Freedom-to-operate searches
• Patent invalidity assessments
• Invention disclosure drafting
• Patent translation quality benchmarks

Each of these areas presents unique challenges that require domain-specific evaluation metrics, which generic AI models and tools have historically failed to address.

Setting a New Standard for AI in IP

Patent Bench represents a milestone in the intersection of artificial intelligence and intellectual property. By introducing a tailored, transparent, and technically sound benchmarking system, Patsnap is not only offering valuable insights for professionals but also setting a new standard for how AI tools are assessed in the legal-tech space. As AI continues to evolve and permeate highly specialized fields, the importance of such benchmarks will only grow. With Patent Bench, Patsnap reinforces its position as a forward-thinking innovator at the forefront of AI-driven IP intelligence.