Harvey vs Legora: pricing, hosting, AI models and German secrecy rules compared (October 2026)

Harvey or Legora: the verdict

Harvey vs Legora: comparison table

Compare all fields side by side, including certifications and hosting →

Pricing: neither vendor publishes one, so read the contract papers before the demo

Harvey's website shows no plans and no prices. Every call to action reads "Request a Demo", and harvey.ai/pricing returns a 404 page. Legora is the same: its calls to action read "Book a demo" and legora.com/pricing returns a 404 page too. Neither site shows a free trial or a free plan. In our directory, 47 legal AI tools are listed, 38 of them sell on request, 9 offer a free trial and 8 have a permanent free plan. Both tools sit in the group that sells by quote.

Several comparison blogs quote per-seat figures, minimum seat counts and median contract values for both vendors. We checked for a vendor document behind those numbers and found none, so they stay out of this page. A figure that cannot be traced to the vendor is a rumor.

What you can read before a demo is the contract set. Harvey publishes its Platform Agreement, Service Terms, Security Addendum, Data Processing Addendum, Data Transfer Addendum and subprocessor list at harvey.ai/legal. Legora publishes an EU data processing agreement, EU general terms and conditions, a specification of the processing, security measures and a pre-approved sub-processor list, with separate sets for the EU, the UK, the US and Australia. Reading these takes an afternoon and tells you more about the commercial relationship than a price range would.

When you ask for a quote, ask for the license unit (named user, seat bundle or usage), the minimum number of seats, which modules are included, what the agent costs to run at volume and how renewal pricing is set. Harvey reports cost reductions of 3 to 5 times against a frontier-model-only setup by routing tasks to cheaper models, which is a claim about its own costs, not a promise about your invoice. For context on the wider market, see our ranking of AI contract review software.

Scale: Harvey states 3,000 customers, and Legora's count rose from 800 to 2,100 in six months

Both vendors publish customer counts, but not in the same unit or on the same date. Harvey's home and company pages state 3,000+ customers ("legal organizations") in 70+ countries, 80+ Am Law 100 firms and 200,000+ professionals, without a date. Its September 9, 2026 announcement says 80 percent of the Am Law 100 use Harvey, together with five of the Fortune 10.

Legora dates every figure. On March 10, 2026 it counted 800 customers. On April 30 it said more than 1,000 organizations and annual recurring revenue above $100 million, a figure it reports itself. On August 26 the count was 1,500, on September 14 and 17 it was 1,800, and on September 28 it was 2,100 firms and legal departments across more than 80 markets. Each number is a floor ("more than"), so treat the trend as the finding: the stated floor rose from 800 to 2,100 in 29 weeks.

On money, the public record is also dated. Harvey announced $550 million at a $15.5 billion valuation on September 9, 2026, co-led by Diffusion and Lightspeed. Legora announced $550 million at $5.55 billion on March 10, extended to $600 million at $5.6 billion post-money on April 30. Bloomberg reported on September 22 that Legora is in talks to raise at least $300 million at about $8.5 billion pre-money, according to people familiar with the matter. That report is unconfirmed. For a buyer, financing says something about staying power and nothing about fit.

Product: Harvey builds around the matter Space, Legora around an operating system and a client Portal

Both products now follow the same pattern: an agent plans and carries out multi-step legal work, and lawyers review the result. Harvey II, introduced on August 18, 2026, makes the matter the starting point. Agents open inside a Space with the documents, parties, tasks, permissions and ethical walls of that matter, and a Memory feature stores your preferences for structure and citation style, which you can view, change or switch off. Legora's Agent sits inside what it calls the agentic operating system (aOS). It plans, executes, reviews and delivers, uses every other Legora tool, and supports Skills that encode a team's way of working.

For bulk review, Harvey offers Vault and a Review Table that its research post describes as working over up to 10,000 documents at a time. Legora's Tabular Review turns thousands of documents into a spreadsheet with each cell cited to its source, and adds real-time collaboration with locked cells and a review mode. Both integrate with Word and Outlook, both have mobile apps, and both connect to document management systems: Harvey lists iManage, NetDocuments, Google Drive and SharePoint, and Legora has a NetDocuments partnership.

The sharper difference is client delivery. Legora's Portal is a branded workspace where a firm shares files and runs its own workflows for clients, with no guest license required. Harvey's Spaces bring firms, clients and co-counsel into a shared workspace with ethical walls. Both do the job. Legora markets it as a product line, Harvey as part of the matter model. Harvey's Command Center adds usage analytics by team, practice group and office, and benchmarks usage against comparable organizations.

Two items need labels. Harvey's Everlaw integration was announced on September 3, 2026 and is expected for joint customers in fall 2026, so it is not yet generally available. Legora's new legal research foundation, with an ontology of law and an AI-native citator, was in limited beta on September 14, 2026 and is planned for general availability in the fourth quarter. Legora's NetDocuments connection through the Model Context Protocol was announced on August 26 with availability to be communicated over time. Its ChatGPT Enterprise plugin, announced on September 17, is available.

Models and data: Harvey names ten AI providers and its own model, Legora names a shorter list

Harvey's subprocessor list, last updated September 21, 2026, names ten AI model and infrastructure providers: Microsoft, OpenAI, Google Cloud, Amazon Web Services, Anthropic, ElevenLabs, DeepL, Mistral AI, Baseten and Fireworks.ai. Three more serve product features: you.com and Parallel Web Systems for web browsing and LexisNexis for its Ask LexisNexis feature. The big four cloud and model vendors are listed with processing in the US, the EU, Switzerland and Australia. Anthropic is listed for the US and Mistral for the EU, each with a footnote that customers who chose other locations may still reach those models through the model hosting platforms of AWS or Google.

Harvey says it contractually guarantees that customer data is not used to train models, requires zero data retention from model providers and keeps each customer's data logically separated. Its August 20, 2026 research post introduces Harvey Tenet, a model post-trained on an open-weight Kimi K3 base together with Fireworks, and states that no customer data was used in that training. The company argues that law firms cannot be locked to one model because of client conflicts, which is why it runs its own agent infrastructure.

Legora's EU sub-processor list, updated July 6, 2026, names eight entries: Microsoft, DeepL (optional), AWS, Google, OpenAI, Intercom, Linkup Labs and turbopuffer. Each is listed with EU or EEA data residency, and the US list adds Exa Labs. Legora's security policy names Azure as its cloud environment and says customer data is hosted in the region specified in the agreement, and that vendors process it in the EU or in a GDPR-compliant way. It states that it will not train or fine-tune any AI model on customer data.

On controls, Harvey lists SAML single sign-on, audit logs, IP allow-listing and data lifecycle management, with customer-set retention and deletion. Legora lists all common single sign-on protocols, a bring-your-own-key option for encryption, access to customer data only with written customer approval, and deletion of all data after contract end, after an export is offered. The Harvey pages we read do not mention customer-managed keys. Both hold ISO/IEC 27001, ISO/IEC 42001 and SOC 2 Type II attestations, and Harvey adds ISO/IEC 27701 and AIUC-1.

Country fit: both opened Munich offices in 2026, and § 43e BRAO sets what a German firm must sign

German lawyers who give an outside provider access to client secrets fall under § 43e of the Federal Lawyers' Act. The provider must be chosen with care, the contract must be in text form and bind the provider to confidentiality with a warning about the criminal consequences, to learning secrets only as far as needed and to rules for any further persons involved. For services provided abroad, the protection of secrets there must be comparable to Germany's. For a service that directly serves a single client matter, the client has to consent. Data protection law applies on top.

The Federal Bar's guidance of December 2024 says it is not conclusively settled whether the data protection level can serve as the yardstick for US providers, and that, where possible, AI providers with server locations in Germany or Europe should be preferred. Neither of these vendors is German. Harvey is a US company that offers processing in the EU and Switzerland, the US or Australia and publishes a Data Transfer Addendum. Legora is a Swedish company whose EU contract set lists EU/EEA residency for its sub-processors. Neither page makes a statement about § 43e itself, so ask for the clause language.

Both are building German teams. Harvey expanded its local German team on July 9, 2025, runs a Munich office opened in March according to its September 24, 2026 release, has a Regional Vice President for DACH, Mihály Gündisch, and on September 24 announced law school partnerships with four German universities. Legora's Regional Vice President for DACH, Björn Franke, has built the region since June 2026, and Legora opened its Munich office on September 29. It reports more than 13,000 legal professionals in Germany, up from about 3,000 at the start of the year, and more than 200 firms and legal departments across Germany, Austria and Switzerland.

On German content, Harvey's Knowledge module lists Otto Schmidt and Wolters Kluwer among its data partners, with access to more than 500 legal sources. Legora's research module lists Wolters Kluwer, EUR-Lex and the German federal law collection among its sources. Test both on your own German questions, because these lists describe access, not quality.

Two more points matter in Germany. Harvey's Command Center shows who uses the tool on what work across teams and offices. Under § 87 paragraph 1 number 6 of the Works Constitution Act (BetrVG), a works council co-determines technical systems meant to monitor employee behavior or performance, so expect that conversation in an in-house department or a firm with a works council. On the EU AI Act, Annex III point 8(a) lists AI used by or for a judicial authority. It does not name a law firm's research tool, but Article 4 on AI literacy has applied to companies using AI systems since February 2, 2025.

Evidence gaps: no independent test has run both products on the same tasks

We found no independent test that runs Harvey and Legora on the same tasks. Harvey created the Legal Agent Benchmark, and Vals AI hosts its leaderboard, updated on September 30, 2026. That leaderboard ranks base AI models from providers such as Anthropic, Google and OpenAI, not the products. It says something about the models that sit under these platforms. It says nothing about which platform gives better answers.

Customer quotes and vendor surveys on both sites are also vendor material. Legora cites a study of 30 customers in 9 countries by an industry analyst, and its April 30 release reports 4.3 saved non-billable hours per lawyer per week among law firms it surveyed. Treat these as indications from the seller. A two-week pilot on your own matters, with the same ten documents and the same ten questions in both tools, is the only test that answers your question.

Who should pick Harvey and who Legora

The bottom line

Treat price as unknown and make the vendors prove everything else. Ask both for a written quote with license unit and seat minimum, read their data processing agreements and sub-processor lists, confirm the processing region in your contract, and run the same ten documents and ten questions through both products. Choose Harvey if matter Spaces, publisher data and model routing matter most to you. Choose Legora if client delivery, your own encryption keys and region-specific contract sets matter most.