PatSnap

What it is

PatSnap is a patent and R&D intelligence platform: a licensed global patent and literature corpus with search, analytics, and now a set of AI agents on top. Jeffrey Tiong founded it in Singapore in 2007, and it reached a $1B valuation in 2021 on a $300M Series E led by SoftBank Vision Fund 2 and Tencent. The vendor claims more than 18,000 customers, 214M+ patents across 174 jurisdictions, and 259M+ scientific literature records, with chemical structure and biosequence databases for life-science and materials teams.

The product line splits into three layers. Analytics is the classic search-and-landscape workbench IP departments have licensed for a decade. Eureka is the AI layer — 23 domain agents on the current catalog, including Novelty Search, FTO Search, Design FTO Search, Competitive Landscape, Technology Roadmap, Technical Disclosure, and Patent Drafting. Open Platform exposes the same data and agents to your own tooling through REST API, MCP servers, embeddable widgets, and packaged Agent Skills. Vertical products (Bio, Chemical, Synapse for biopharma, Hiro LS) sit beside them.

That positioning matters for anyone reading our Patlytics, Solve Intelligence, or DeepIP pages. Those tools produce work product — claim charts, drafted applications. PatSnap is the data and search layer underneath that work: where the prior-art set, the landscape, and the FTO screen come from.

Why it shows up in Legal Ops stacks

• The corpus is the moat, not the chat box. Full-text patents with normalized assignees, legal status, and family data across 174 jurisdictions, plus literature, chemical structures, and sequences in the same index. An AI-native drafting startup cannot replicate a licensed data estate of that size, and an FTO opinion is only as good as the set it searched.
• Agents are scoped to named IP jobs. FTO Search and Novelty Search return ranked references with reasoning rather than a keyword hit list, and Design FTO Search covers design patents, which most AI entrants skip. The vendor claims an 85% hit rate in novelty search ahead of ChatGPT and Claude — a vendor benchmark on a vendor-chosen set, so treat it as a claim to test.
• The developer surface is unusually open for an incumbent. The vendor advertises 20+ MCP servers for Claude Desktop, Cursor, and other MCP clients, 300+ data APIs, and 60+ agent APIs, plus LangChain and CrewAI integration. A legal-ops engineer can start on the free Starter tier and put patent search inside an internal agent the same afternoon, without a separate data-license negotiation.
• R&D can use it without the IP team in the loop. Eureka is sold to engineers as much as to counsel: TRIZ problem solving, technology roadmaps, and invention disclosure drafting. That widens adoption, and it also widens the confidentiality surface (see watch-outs).

Pricing reality

PatSnap publishes more than most of the segment, but only at the edges. Eureka Pro lists at $100 per month ($1,080 per year) after a 14-day trial with 10,000 credits; Life Sciences Pro lists at $400 per month ($3,600 per year). Open Platform is credit-based at $100 per 10,000 credits (one credit is one cent), with a free Starter tier of 10,000 credits for 90 days at 10 requests per minute, Pro at 60 queries per minute, and custom Enterprise above 120.

The core Analytics license — what an IP department actually buys — is quote-only. Buyer-reported ranges put entry contracts at roughly $15K–30K per year for a few seats, with Insights, Discovery, or the chemistry and biology modules pushing toward $50K and beyond; multi-seat enterprise deals run into six figures. Those are reported ranges, not published prices. Two levers that reportedly move the number: a written competing quote from Clarivate or Questel, and buying one or two power seats instead of licensing occasional users.

Put three things in the order form: which modules each seat can access, the credit burn rate for each agent you plan to automate, and whether Open Platform credits are pooled with or separate from the Analytics contract.

Best for

Corporate IP and R&D teams at companies filing or clearing dozens of inventions a year — especially in chemistry, materials, and life sciences — who need landscape, novelty, and FTO searches on a defensible licensed corpus, and want that corpus reachable from their own agents over MCP or API.

Not for you if your bottleneck is drafting speed or claim-chart volume: PatSnap’s Patent Drafting agent exists, but drafting-first tools sit closer to the attorney’s work. Not for you if you need docketing, annuities, or IP spend management — that is Anaqua’s job. And not for a solo practitioner who runs a few searches a quarter; Lens.org and Google Patents are free.

Alternatives — and when to pick them instead

• Clarivate (Derwent Innovation) — the other incumbent by market share, with hand-curated Derwent abstracts and family data that litigators trust. Pick it when the search record may be scrutinized in litigation or an opposition, and curated abstracts outweigh agent tooling.
• Questel (Orbit Intelligence) — the third major data incumbent, strong in Europe and bundled with IP management services. Pick it when you want search, docketing support, and renewals from one vendor.
• Patlytics — the fastest-growing AI-native entrant on the analysis side, $40M Series B in April 2026. Pick it when the deliverable is an evidence-of-use or invalidity claim chart, not a search report.
• Solve Intelligence or DeepIP — pick one when the workload is drafting and prosecution and search is an input rather than the product.
• Lens.org, Google Patents, EPO OPS — free. Pick them when volume is low and no one needs a vendor’s coverage guarantee.

Watch-outs

• Every performance claim is vendor-sourced. The 85% novelty hit rate, “75% faster at 25% less cost,” and the 18,000-customer count are PatSnap marketing. Guard: run the POC on two or three inventions where you already know the closest prior art, and score whether each agent surfaces it in its top ten results.
• SOC 2 is Type 1, not Type 2. The trust page lists ISO 27001:2022 and SOC 2 Type 1 — a point-in-time design attestation, not an operating-effectiveness audit. Guard: request the report and bridge letter, and put data residency, model-training exclusion, and the subprocessor list in the DPA before R&D starts pasting unfiled invention disclosures into Eureka.
• Agent Skills are still beta. The vendor’s own April 2026 post says Agent Skills were still rolling out in Q2 2026. Guard: build production automations on the REST API or MCP servers, and pin the agent versions you depend on in writing.
• Credits make automation cost elastic. Agent calls cost more credits than basic retrieval, and an internal bot that runs FTO screens on every engineering ticket can burn a year’s allowance in weeks. Guard: set per-key rate limits and a monthly credit alert before you connect any agent to a workflow that runs unattended.
• Search results are not an opinion. An FTO Search output is a ranked reference set, not clearance. Guard: keep a registered practitioner’s sign-off on every FTO conclusion, and archive the query, corpus date, and results so the search record survives a willfulness challenge.