Guides
What Legal AI Really Costs in 2026: Harvey, Legora, CoCounsel, Claude and ChatGPT
Legal AI cost as of September 2026, from €39 to $1,200 per seat per month, with Harvey, Legora, CoCounsel, Claude and ChatGPT prices and worked budgets for 5 and 50 lawyers.
On this page
- The price list nobody publishes
- General-purpose assistants: what a seat costs
- Legal verticals: from €39 to $1,200 a month
- Seat pricing versus consumption pricing
- A cost model for a five-lawyer firm
- A cost model for a 50-lawyer firm
- What "custom pricing" hides: ten questions
- Passing costs on to clients
- Where this leaves you
Legal AI in September 2026 costs anywhere from €39 to about $1,200 per lawyer per month, a thirty-fold spread that vendors rarely put on one page. A frontier assistant on a business plan is $20–30 a seat. Legal verticals start at $104 (CoCounsel) and climb to Harvey's estimated $1,000–1,200 with a 20-seat minimum. This guide puts the numbers side by side and models two firm sizes.
The price list nobody publishes
Only a handful of legal AI vendors print a price. Thomson Reuters publishes a range for CoCounsel Legal, Clio publishes $199 for Clio Work, Spellbook and GC AI publish per-user rates. Harvey, Legora and LexisNexis do not. Everything below is stated as of 18 September 2026; figures marked "reported" come from secondary sources cited in the Harvey review and elsewhere on this site, not from a rate card.
Two events set the tone for 2026 pricing. Harvey raised $550m at a $15.5bn valuation on 9 September 2026 with ARR above $400m (LawSites); at roughly 39 times revenue, the seat price is not coming down. In the other direction, Anthropic's Claude for Legal plugins (12 May 2026) are open source and free to paid Claude customers, and Clio's $199 Clio Work put the vLex corpus into solo hands on 21 April 2026 (LawSites). The floor dropped while the ceiling rose.
General-purpose assistants: what a seat costs
| Tool | Plan | Price (Sep 2026) | Trains on your data? | Minimum |
|---|---|---|---|---|
| Claude | Pro | $20/month ($17 annual) | Only if you opt in (5-year retention if you do) | None |
| Claude | Max | $100 or $200/month | Only if you opt in | None |
| Claude | Team | $25 standard / $125 premium per seat | Never | 5 seats |
| Claude | Enterprise | Custom | Never | Negotiated |
| ChatGPT | Plus | $20/month | Yes unless you opt out | None |
| ChatGPT | Pro | $100 or $200/month | Yes unless you opt out | None |
| ChatGPT | Business | $25/seat monthly or $20 annual (Premium $125/$100) | No | Negotiated |
| ChatGPT | Enterprise | Custom | No | Negotiated |
| Gemini | Google AI Pro / Ultra | Reported at roughly $20 and $250/month; personal accounts only | Consumer chats used for training unless Keep Activity is off; a subset read by humans | None |
| Gemini | Workspace add-on / Gemini Enterprise | Add-on pricing; Enterprise pay-as-you-go since 1 Aug 2026 | No | Negotiated |
| Microsoft 365 Copilot | Add-on | $30/user/month, annual commitment | No (runs under the Microsoft DPA) | None |
| Perplexity | Pro / Max | $20 / $200/month | Consumer: toggle in settings | None |
| Perplexity | Enterprise Pro / Max | $40 / $325 per seat | Never by default | Negotiated |
| Mistral Vibe | Pro | $14.99/month | See terms | None |
Two rules of thumb. First, the seat price is the small part: the data terms decide whether you may use the tool for client work at all, which is why the confidentiality checklist matters more than this table. Consumer tiers train unless you opt out; business and API tiers do not. Second, the strongest models cost more in the API and are absent from free tiers: Claude Fable 5.1 is $10 in and $50 out per million tokens and is not on the Free plan; GPT-6 Astra is also $10/$50; Opus 5 is $5/$25; Sonnet 5 $2/$10; Gemini 3.1 Pro $2/$12 (Anthropic's model overview and ChatGPT pricing summary). A million tokens is about 555,000 words, so a 300-page data room read once through Opus 5 costs under a dollar in input tokens. Tokens are cheap; seats and people are not.
Legal verticals: from €39 to $1,200 a month
| Product | Published or reported price | Minimum | Notes |
|---|---|---|---|
| AI:ssociate (Austria) | From €39/user/month | None | ÖRAK partner; three free months via the bar |
| Clio Manage AI tiers | $49–$149/user/month | None | Practice-management bundle |
| CoCounsel Legal (Thomson Reuters) | $104–$639/user/month (published); solo Westlaw Advantage + CoCounsel Essentials $639.20 | None | Next-generation version GA in the US 20 Aug 2026; NPS 77 |
| Spellbook | About $180/month (reported range $99–$350) | None | Word add-in; ~4,000 customers |
| Clio Work | $199/user/month | None | Built on vLex Vincent; 1bn+ documents, 110+ jurisdictions |
| GC AI | $500/user/month | None | In-house focus |
| LegalOn | About $550/month (reported) | Not stated | Five agents launched Feb 2026 |
| Blue J (tax) | About $1,500/seat/year | Not stated | Tax research |
| Legora | From about $3,000/user/year (reported) | 10 seats (reported) | $5.6bn valuation; 1,480+ customers |
| Harvey | About $1,000–1,200/seat/month (reported) plus consumption | 20 seats (reported) | Token-based billing reported Sep 2026 |
| Lexis+ with Protégé | Not published | Negotiated | US GA 24 Feb 2026 |
| Astra for Law (OpenAI) | Not published; announced 17 Sep 2026 | Unknown | Too new to price |
| Claude for Legal plugins | No extra charge for paid Claude plans | Claude plan | Apache 2.0, open source |
| Microsoft Legal Agent for Word | Included in the $30 Copilot add-on | Copilot licence | Frontier preview; needs Anthropic enabled as subprocessor |
The Harvey and Legora figures come from secondary sources (a GC AI review and Spellbook comparison pages), and the Harvey vs Legora comparison discusses how they bundle Vault, Knowledge and agent runs. CoCounsel's range is the only published enterprise-grade price, which is itself a competitive statement. ILTA's 2026 Technology Survey (14 September 2026) shows why the top of the table can hold its price: Harvey is used by 43% of responding firms and 84% of firms with 700 or more lawyers (LawSites).
Seat pricing versus consumption pricing
Until 2025 almost every legal AI product was sold per seat. Agents changed that, because an agent that reads 10,000 documents against 100 questions overnight (Thomson Reuters' Tabular Analysis) or answers counterparty redlines before you arrive (Legora's aOS) burns tokens without a human in the chair. Three 2026 data points show where this is going:
- Harvey's pricing is described as seat plus consumption, with reports of token-based billing in September 2026.
- Microsoft's Copilot Cowork reached general availability in June 2026 on usage-based billing rather than per seat.
- Anthropic's Managed Agents (public beta 8 April 2026) cost $0.08 per session-hour plus tokens; Relativity aiR (16 September 2026) folded its agentic features into platform pricing instead.
Consumption pricing rewards light users and punishes exactly the workflows agents are sold for. The sensible negotiating position for a firm is a seat price with a documented monthly token allowance per seat, a stated overage rate, and a dashboard you can see. If a vendor cannot show you last month's consumption per user during a pilot, you cannot forecast next year's invoice. The six-week pilot playbook includes a consumption log for this reason.
A cost model for a five-lawyer firm
Assume five fee-earners, one assistant, Microsoft 365 already in place. Figures are monthly list prices as of September 2026, before tax.
| Setup | Components | Monthly | Annual |
|---|---|---|---|
| Lean general | Claude Team standard (5 × $25) or ChatGPT Business (5 × $25); Copilot for two people (2 × $30) | $185 | $2,220 |
| Lean plus Word | Lean general plus Spellbook for two drafters (2 × ~$180) | $545 | $6,540 |
| Small-firm vertical | Clio Work (5 × $199) plus Claude Team standard (5 × $25) | $1,120 | $13,440 |
| Research incumbent | CoCounsel Essentials with Westlaw Advantage at the solo bundle price (5 × $639.20) plus Claude Team | $3,321 | $39,852 |
| Harvey | Not available: 20-seat minimum implies $20,000–24,000/month even for five users (reported) | n/a | n/a |
| Legora | 10-seat minimum implies about $2,500/month for five users (reported) | $2,500 | $30,000 |
A solo or five-lawyer firm therefore has a realistic band of $2,000 to $40,000 a year. The small-firm guide argues that the lean setups plus disciplined verification cover most work, and the question page on whether Harvey is worth it for a small firm reaches the same conclusion from the other side. Clio's 2025 Legal Trends Report found that legal-specific tool use among lawyers fell to 40% from 58% as people defaulted to general chatbots, which is the market voting with its feet on this table.
A cost model for a 50-lawyer firm
Fifty fee-earners, an IT function, iManage or NetDocuments, and a partner who has been asked "what is our AI budget". Seat licences only, annualised, September 2026 list or reported prices:
| Option | Calculation | Annual seats | What is not included |
|---|---|---|---|
| Copilot everywhere | 50 × $30 × 12 | $18,000 | Legal Agent limited to US/Windows preview |
| Claude Team premium | 50 × $125 × 12 | $75,000 | Enterprise features, ZDR (Team chat is not ZDR-eligible) |
| Clio Work | 50 × $199 × 12 | $119,400 | Transactional agent depth |
| Legora | 50 × ~$3,000 | $150,000 (reported entry price) | Consumption, onboarding |
| CoCounsel mid-tier | 50 × ~$300 × 12 | $180,000 (range $62,400–$383,400) | Westlaw content tier |
| GC AI | 50 × $500 × 12 | $300,000 | Aimed at in-house |
| Harvey | 50 × ~$1,100 × 12 | About $660,000 (range $600,000–$720,000) | Token consumption, Vault storage, implementation |
Now the context. Citi's 2026 Hildebrandt Client Advisory reports that law firms spent 2.4% of revenue on technology in 2024. A 50-lawyer firm at $500,000 revenue per lawyer has $25m of revenue and, on that average, a $600,000 technology budget for everything: DMS, practice management, security, laptops. Harvey at full deployment would consume it entirely. That is why most firms at this size run a layered stack (Copilot for everyone, a frontier model for most, a vertical for a deal or litigation team), and why the decision framework starts from use cases rather than vendors. The build-or-buy guide covers what happens above 200 lawyers, where Latham & Watkins and Kirkland & Ellis have started buying their own GPUs.
Add three lines to any budget that seat tables omit. Implementation and DMS connection (iManage's MCP server shipped 14 May 2026, but someone still configures permissions). Training, which the AI-literacy guide shows is now a documented duty in the EU. And the verification tax: Harvey's own LAB benchmark scored the best model at 13.3% on the strict all-criteria-pass metric on 18 September 2026, so every agent output still needs a lawyer's hour.
What "custom pricing" hides: ten questions
Ask these before signing anything priced "on request":
- What is the minimum seat count and minimum term, and do unused seats roll over?
- Is there any consumption element (tokens, agent runs, documents processed, storage), and what is the overage rate?
- Which models are included, and is there a surcharge for the newest (Fable 5.1, GPT-6 Astra) or for fast modes (Claude's fast mode costs double in the API)?
- What are the implementation, onboarding and DMS-connector fees?
- Is training included, and does it produce the record Article 4 of the EU AI Act expects?
- What is the data retention period, is zero data retention available, and which models are excluded from it (Anthropic's Fable and Mythos models require 30-day retention)?
- What is the annual price escalator?
- Can you export prompts, playbooks, agents and Vault contents on exit, and in what format?
- Are guest or per-matter accounts available for clients and co-counsel, and how are they priced?
- What happens to the price if the vendor is acquired or changes its model stack?
Question 8 is not theoretical. OpenAI announced on 3 June 2026 that its Agent Builder shuts on 30 November 2026, Relay.app closed to paying customers on 14 September 2026, and the ChatGPT Atlas browser lasted 292 days. Contract for data export from day one, and put the answers through the vendor due diligence questions before procurement does.
Passing costs on to clients
Whatever you pay, the ethics rules limit what you recover. ABA Formal Opinion 512 (July 2024) treats general AI subscriptions as overhead, like Westlaw, and allows billing actual time spent, not time saved. California's Board of Trustees guidance of 14 May 2026 says the same and adds that AI costs may not be marked up without informed written consent. Texas Opinion 705 (February 2025) forbids charging for time "saved" by AI; Virginia LEO 1901 (November 2025) confirms a reasonable value-based fee need not be cut because AI made the work faster. The billing clients for AI page works through the cases, and the billable hour guide covers the commercial side: who keeps the dividend.
Where this leaves you
Budget in three layers. A business-tier general assistant for every fee-earner ($20–30 a seat) is now the baseline, and the AI Lab for Lawyers course is built on the assumption that most of the value for a working lawyer sits in that layer, using browser-based tools and no installs. A legal vertical is a second layer for a team with a repeatable, high-volume task (diligence, NDA review, litigation research), priced per seat with a usage cap. A third layer, own infrastructure, is for firms with a nine-figure technology budget or a confidentiality need that no cloud vendor can meet. Re-check every number above quarterly; in a market where a $20 plugin knocked 14% off RELX in a day (3 February 2026), price lists do not stay still.
Frequently asked questions
How much does Harvey AI cost per user?
Harvey does not publish prices. Secondary sources in 2026 put it at roughly $1,000 to $1,200 per seat per month with a 20-seat minimum, plus consumption elements, and there are reports of token-based billing. Treat any figure as an estimate until you have a written quote, and ask what the minimum term and overage rules are.
What is the cheapest legal AI tool for a small firm in 2026?
Austria's AI:ssociate starts at €39 per user per month, with three free months through the Austrian bar. In the US, Clio Work costs $199 per user per month and CoCounsel starts at $104. A business-tier Claude or ChatGPT plan at $25 per seat covers most drafting and analysis for a solo or small firm.
Is seat pricing or consumption pricing better for a law firm?
Seat pricing is predictable and easy to budget; consumption pricing is cheaper for light users and punishing for heavy agentic workflows that burn tokens overnight. Most firms should insist on seat pricing with a documented usage cap, and ask any vendor moving to token metering for a worked example based on your own matters.
Can a law firm bill clients for AI subscription costs?
Generally not as a separate charge. ABA Formal Opinion 512 and the California State Bar's 2026 guidance treat general AI subscriptions as overhead, like a research database. A matter-specific tool used at a client's request may be passed through at cost, but a mark-up needs informed written consent in California.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Questions
How Much Does Harvey AI Cost? Pricing, Minimums and Alternatives
Harvey AI costs an estimated $1,000–1,200 per seat per month with a 20-seat minimum as of September 2026. What you get, how to negotiate and cheaper options.
6 min read
Tools
Harvey AI in 2026: What It Does, What It Costs and Who Uses It
Harvey AI reviewed as of September 2026. The $15.5bn valuation, Harvey II, Tenet, 500-plus agents, who uses it, estimated seat prices and its own benchmark.
9 min read
Tools
CoCounsel Legal (Thomson Reuters) Review: Westlaw-Grounded Agents
CoCounsel Legal reviewed as of September 2026. The Claude-built rebuild, Brief Builder, Tabular Analysis, Deep Research Verify, $104 to $639 pricing and grounding.
8 min read
Regulation & Ethics
Billing Clients for AI: What You Can Charge and What You Cannot
Lawyers may bill actual time spent using and checking AI or a fixed fee agreed in advance, but not saved hours, subscriptions or marked-up AI costs. Worked numbers inside.
10 min read
Comparisons
Harvey vs Legora: The Two Legal AI Platforms Compared
Harvey vs Legora in September 2026: $15.5bn vs $5.6bn valuations, ARR, clients, Workflow Agents vs Agentic OS, models, pricing minimums, geography and who should pick which.
10 min read
Questions
Is Harvey Worth It for a Small Law Firm?
Harvey is usually not worth it for a small law firm: a reported 20-seat minimum at about $1,000–1,200 a seat. When it is, and what six lawyers should buy instead in 2026.
6 min read