Align Research Prices Legal AI at $100 a Job

Align Research, a new legal AI tool from former Williams & Connolly partner Sam Davidoff, charges a flat $100 per research job instead of a monthly seat. He opened general availability on August 3, 2026, with three free jobs a month and no subscription.

Davidoff sells the product, so he is arguing his own book. The older legal-research meter still sits on library price guides, and the token invoices under every “unlimited” seat have not gone away.

A Few Hours Later, Just the Cases

Align is a Washington, D.C. litigation shop Davidoff built after leaving Williams & Connolly. Align Research takes a question in plain language, works for a few hours, and emails back a binder of real court opinions with the passages it thinks matter already marked. There is no memo and no generated answer. The company says the system retrieves rather than writes, which is how it claims it cannot invent a case.

It’s not telling you anything, it’s just showing you. If all the AI is doing is essentially pointing you to pre-existing things, as opposed to writing some narrative for you, it actually can’t hallucinate. It could miss something, it could highlight something that you don’t agree with, but it can’t make something up.

Sam Davidoff, founder and CEO, Align

He said the research task he assigned as a partner was rarely “answer this question.” It was “get me the cases,” including every district court that had granted a motion to dismiss on a given theory. Most legal AI products now return a synthesized answer. He wanted the stack of opinions an associate used to walk into a partner’s office with.

Only Real Opinions Come Back

Coverage is U.S. federal and state case law drawn from the Free Law Project’s CourtListener database, which lists 8.3 million precedential opinions across 472 jurisdictions. Align is building extra case collections of its own and says it will contribute those opinions back, because Davidoff does not want the raw corpus to be the product. The company also says the product is SOC 2 Type II certified and already used inside AmLaw 100 firms.

A user cannot chat with a finished job. A new question means a new job, or the lawyer downloads the PDFs and feeds them to some other tool. That is slower than a chatbot, and it is the point. The machine does the uninteresting pass through thousands of cases. The lawyer still reads the few dozen that matter.

HOW A JOB MOVES THROUGH ALIGN

• Issue split: One model reads the question and breaks out the legal issues to research.
• Search plan: Another model decides which queries to run against the opinion database.
• Read and sort: Further models read retrieved cases, classify them, and score relevance.
• Loop check: The pipeline decides whether it still needs more searching before it stops.
• Highlight pass: A last stage marks the passages that will appear in the binder.

That chain is why a job can run for a few hours. Davidoff likens it to a junior on staff: you assign the work, go do something else, and wait for the email.

$100 a Job, Three Free Each Month

There is no annual database rental and no seat minimum. Align’s launch notice puts a fixed $100 cost per research job on the matter, with three free jobs each month that do not expire. Davidoff said the firm is bootstrapped, so every search costs real money in model tokens, and he wants that cost visible.

He expects firms to pass the fee through. A task that used to eat a couple of associate hours at a couple of hundred dollars an hour becomes a $100 line item. Two hours at $200 is $400. The $100 job is $300 less than that floor, before anyone counts the partner time that used to wait on the memo.

WHAT FIRMS ARE ASKED TO PAY

Twenty Harvey seats at $1,200 a month would be $288,000 a year before training, implementation, or a Lexis add-on. Align’s paid meter starts after the three free jobs. Those are different products aimed at different workflows, and the invoice shapes are the comparison that matters.

Token Prices the Seat Licenses Hide

Davidoff’s claim is not that language models are expensive in the abstract. It is that every legal AI vendor still pays Anthropic, OpenAI, or Google by the token, and an agent that reads thousands of opinions for hours multiplies that bill. Seat licenses hide the meter from the law firm. They do not delete it.

I don’t think it can last. These token costs are real. The amount of money that every legal tech AI provider has to pay to Anthropic, OpenAI, Gemini, that’s real money and it’s per token.

Sam Davidoff, founder and CEO, Align

OPENAI RATES UNDER THE HOOD

• Flagship promo: GPT-5.6 Sol at $4 per million input tokens and $20 per million output tokens, at least through November 21, 2026.
• Mid tier: GPT-5.6 Terra is $2 input and $12 output per million tokens.
• Volume tier: GPT-5.6 Luna is $0.20 input and $1.20 output per million tokens.
• Top end: GPT-6 Astra is $10 input and $50 output per million tokens, and web search is $10 per 1,000 calls.

A chatbot answer can live on Luna money. A multi-step research agent that loops, rereads, and calls search is a different animal, and output tokens cost several times input. Harvey’s own help pages say it routes work across Anthropic, OpenAI, and Google models, so the same rate card sits under the product that sells seats as if usage were free.

The cheap-Claude counter is already in the market. A $20 generalist account plus plugins can draft a deposition summary a paralegal used to spend a weekend on, and Microsoft has put a legal agent inside Word on Copilot plans many firms already buy. That pressure is real. It does not make an hours-long case-retrieval run free. Inference cost per task is still where margin lives, even when the client invoice says “unlimited.”

Harvey’s $11 Billion Seat Wager

The other side of the bet has much more money. On March 25, 2026, Harvey announced a $200 million round at $11 billion, co-led by GIC, Singapore’s sovereign wealth fund, and Sequoia, with Andreessen Horowitz, Coatue, Conviction Partners, Elad Gil, Evantic, and Kleiner Perkins in the round. The cash is meant to grow more than 25,000 custom agents customers already run on the platform.

Harvey does not publish a price list. Buyer estimates circulating through 2026 put mid-market seats in a $1,000 to $2,000 band per user per month, usually on annual terms with a floor in the low dozens of seats. Some firm owners say they already pay less than that sheet and that the product earns its keep in the first hour. Both can be true: unpublished enterprise pricing is a negotiation, and the meter is still running on Harvey’s side of the API.

THE 2026 MONEY CALENDAR

1. March 25, 2026: Harvey raises $200 million at $11 billion to scale agents inside law firms and legal departments.
2. July 28, 2026: Growth Equity at Goldman Sachs Alternatives and J.P. Morgan’s Growth Equity Partners invest, after a quarter with over $100 million in ARR added.
3. August 3, 2026: Align Research goes generally available at $100 a job, with three free jobs a month and no seat contract.

Harvey CEO Winston Weinberg said the bank money was about bringing “marquee investors” in for the next stage of growth. That is a vote that the seat model still sells. It is also a reminder that the subsidy, if it exists, is being funded by growth equity, not by tokens that have become too cheap to meter.

Westlaw Still Suggests $99 a Search

Law firms have already lived this cycle. Lexis grew up charging by the search. Westlaw leaned on flat subscriptions for heavy users. Neither published a clean public rate card for large firms, and dual subscriptions became insurance against a missed case. When clients pushed back, firms tried to recover the cost on the bill, then watched recovery collapse and folded research into overhead.

The meter never fully left. The Franklin County Law Library’s cost guide, updated July 22, 2026, still describes the out-of-plan warning: open that document and you pay extra. Westlaw still tells firms they may bill $99 per Westlaw search to a client under “Predictable Pricing,” with the clicked documents included unless they sit outside the plan. Lexis still points to $60 a search plus a charge for each document. Not every firm passes those costs through. Plenty now treat research as overhead, or bill only the overages.

That is the historical joke sitting under Align’s price. One hundred dollars a job is almost the same number Westlaw already asks firms to put on a client invoice for a single search. The difference is that Align’s fee is the whole product, not an out-of-plan penalty on top of a six-figure database contract. Davidoff said he is not trying to kill Westlaw or Lexis. “They have amazing databases that they’ve built up.” He wants the hours spent assembling the first pile of cases.

Who Pays When Research Hits the Meter

Clients have been refusing associate research hours, and some have started refusing basic case-law search charges too. A $100 task fee is easier to defend on a bill than three silent hours at a junior rate, because the unit is the job, not the person. For in-house teams, a known per-task price is also a number they can benchmark across matters, which a blended seat license is not.

The hidden party in the seat era is the associate whose prompts never hit a ledger. Unlimited seats encourage everyone with a login to throw long documents at the model, because the firm already paid. A job fee forces someone to decide whether the question is worth $100. That is uglier. It is also how firms used to police Westlaw IDs when connect-time was the bill.

The risk is the same one that killed research recovery the last time. If every AI pass becomes a line item, sophisticated clients will cap it, or demand the tool be treated as overhead the way they demanded Westlaw be treated as overhead. If firms swallow the cost, the $100 job becomes another internal budget fight, and the associate goes back to the “free” seat until that vendor’s token bill shows up in next year’s renewal.

Some practice groups will keep paying for Harvey-class seats because the workflow is diligence, drafting, and multi-step agents, not a pile of opinions. Align does not claim to do that work. It claims to return cases. The market now has both invoices on the table: an annual seat that pretends usage is free, and a $100 job that admits it is not.

OpenAI’s Sol promo holds at least through November 21, 2026, after which the published rate is higher. The token bill does not care which column of the law firm’s budget it lands in.

Frequently Asked Questions

Does Align Research replace Westlaw or Lexis?

Davidoff said he is not gunning for either database, and he called their collections amazing. Align Research is built to assemble the first set of on-point opinions, including a sweep of district courts on a motion theory, then stop. Citators, treatises, dockets, and the rest of a paid research plan stay where they are.

What happens if the results raise a new question?

There is no follow-up chat on a finished job. The lawyer runs another $100 job, uses one of the three free jobs, or downloads the marked opinions as PDFs and loads them into a different AI tool for interrogation. Each new legal issue is a new task, which is how the price stays attached to the work.

Where does Align get the cases it returns?

The public corpus is CourtListener, a Free Law Project site with 8.3 million precedential opinions and 7,255 cases added in a recent ten-day window. Align is also building its own case-law stores and says those opinions will be contributed back, so the underlying corpus is not the lock-in.

Can the tool still miss a case a lawyer needed?

Yes. Davidoff said the system may still “hallucinate” under the hood, but the user only ever sees actual court opinions with highlights. The failure mode he accepts is a missed case or a highlight a lawyer disagrees with, not a fabricated citation in a memo.

Disclaimer: This article is news reporting and analysis of legal-technology products and prices. It is informational only and is not legal advice, a solicitation to buy software, or a recommendation of any vendor. Law firms, legal departments, and other buyers should review contracts, security terms, and matter-billing rules with their own counsel, procurement team, and accountant before changing research tools or passing AI fees through to clients. Figures and product terms reflect vendor pages, library guides, and public statements as of the dates named above and can change with the next rate card or contract cycle.