If you budget a review matter on headline per-GB rates, you are comparing the wrong number: on a 500,000-document review, hosting is a rounding error next to each contract's charge for AI first-pass review. As of November 15, 2026, Relativity includes aiR for Review, aiR for Privilege and aiR for Case Strategy in committed data volume, so the AI line is zero if you commit. Everlaw includes single-document AI in its per-GB rate but still meters batch coding suggestions in credits. DISCO's per-GB price bundles its Cecilia AI features, yet Auto Review, the product that actually codes documents, carries its own per-document charge.
The recommendation: if you run more than a few review matters a year, buy RelativityOne (now sold as Relativity aiR) on a one- or three-year commitment and get the AI inclusion in writing. Pick Everlaw if your matters are small, your team is lean and you value unlimited users and a single bill more than free bulk AI. Skip DISCO for AI-led first-pass review today: its review engine is priced outside the bundle it advertises, and neither number is published.
| Relativity (aiR, formerly RelativityOne) | Everlaw | DISCO | |
|---|---|---|---|
| Billing unit | Per GB per month; pay as you go, or 1- or 3-year commit | Per GB per month, unlimited users | Per GB of processed data, one line item |
| GenAI first-pass review | aiR for Review included with committed data volume | Single-document suggestions included; batch runs use credits | Auto Review: separate per-document charge |
| GenAI privilege review | aiR for Privilege included | Not separately listed | Not separately listed |
| Published commercial price | None ("Request Pricing") | None (quote) | None (quote) |
| Public reference rate | Epiq UK partner card: £7.00/GB/month review hosting | GSA federal list: $25.24/GB/month active data | None found |
| Dormant matter storage | Cold storage tier; Epiq dormant £1.50/GB/month | Suspended data; GSA $3.50/GB/month | Vault rate not published |
| Who should NOT buy | Firms with one or two matters a year that cannot commit volume | High-volume reviewers who want zero metered AI | Buyers who need a fixed AI review cost before the matter starts |
| Verdict | Lowest marginal AI cost for repeat buyers | Safest for small, predictable matters | Loser for AI-led first-pass review |
Prices checked on each vendor's live page and the named public price lists on November 15, 2026.
What the Winter 2026 Survey Says Generative AI Review Costs
Generative AI review is most often priced at $0.11 to $0.50 per document, below the $0.50 to $1.00 most buyers pay for human first-pass review, and about a third of respondents answered "do not know" or "not applicable." Those figures come from the Winter 2026 eDiscovery Pricing Survey run by ComplexDiscovery OÜ with EDRM: 53 practitioners, collected December 2025 to February 2026, 92.5% of them in the US. Treat it as self-reported ranges, not invoices.
The per-document answers (question 19) split like this: 20.8% reported $0.26 to $0.50, 15.1% reported $0.11 to $0.25, 15.1% reported $0.05 to $0.10, 7.5% above $0.50 and 5.7% below $0.05. The largest single answer was "do not know/not applicable," at 35.8%.
The billing model has not settled either. Hybrid and per-document pricing tied at 28.3% each, per-GB took 11.3%, and per-token and flat subscription took 5.7% each. The survey's author called the hybrid/per-document tie "the defining finding."
For the human baseline, the document review section of the same survey puts remote managed review most often at $0.50 to $1.00 per document (28.3%), with 18.9% paying above $1.00. Hourly remote reviewers cluster at $25 to $40 (41.5%).
The answer buyers overlook is question 22: what happens to documents that fail or need special handling in AI review. 18.9% route them to manual review at standard rates, 17.0% said it depends, 9.4% bill extra processing time and only 9.4% include them in the base price. That exception pile is billed at human rates, and none of the three vendors' pricing pages addresses it.
What Relativity Actually Includes Now
Relativity's AI review tools cost nothing extra for customers with committed data volume; the conflicting reports mostly describe contracts signed before the change or data bought pay as you go. Relativity announced at Relativity Fest on October 8, 2025 that aiR for Review and aiR for Privilege would become standard in RelativityOne. Before that they were separately priced, which is why older third-party pages still quote per-document aiR charges. Relativity's July 1, 2026 release added aiR for Case Strategy to "the RelativityOne rate," and in September 2026 the platform itself was renamed Relativity aiR.
Read the live pricing page closely, because the wording settles the dispute. It offers pay as you go or a one- or three-year commitment, and says Assist, aiR Review, Privilege and Case Strategy "are all included with your committed data volume." It does not say they are included for pay-as-you-go data. It also lists add-ons "licensed separately from your data volume commitment": Data Breach Response, Translate, Relativity FOIA and Legal Hold. Relativity publishes no dollar figure for any tier and no statement on how AI usage is metered or capped.
Three practical consequences:
- If your contract predates the inclusion, you may still have a per-document aiR line. Ask your account team to move you, in writing.
- If you buy through a partner on a transactional basis, ask whether the partner's Relativity subscription is a committed one and whether aiR runs are passed to you at zero.
- Relativity Server (on-premises) is not mentioned in any of the inclusion announcements. Assume it is excluded until told otherwise.
Partners: no markup on Relativity, but they sell their own AI
Relativity's CEO Phil Saunders told Above the Law in October 2025 that partners cannot mark up a Relativity product. Some partners still sell AI products of their own.
The clearest public example is Epiq's UK G-Cloud 14 price list, valid from March 5, 2026. It prices RelativityOne review hosting at £7.00/GB/month, ECA/repository hosting at £2.50/GB/month (capped at 50 million documents and 5 concurrent users) and dormant hosting with no user access at £1.50/GB/month. User fees are included. All-in processing runs £45.50/GB at the smallest volume tier, falling to £27.65/GB above 1 TB. The same list sells Epiq AI Review plus Epiq Assist at £0.37 per document, a one-time fee "applied to ALL documents ingested to an Epiq AI collection." Because it is charged on ingestion, every document in the collection pays it, including the ones nobody reviews.
Other partners went the opposite way. Cimplifi announced "all-in unlimited" per-GB pricing for aiR for Review and aiR for Privilege on October 9, 2025, with its CEO blaming "unpredictable pricing tied to document volume" for slow AI adoption.
Who should not buy Relativity: a firm with one or two matters a year that cannot commit volume. Pay as you go leaves the AI inclusion ambiguous, and you will end up buying through a partner whose rate card decides the real price.
What Everlaw Charges for AI, and What It Leaves Out
Everlaw folds single-document AI and most drafting tools into its per-GB rate, and charges credits for batch runs. At its October 2025 summit, Everlaw said Review Assistant on single documents, Writing Assistant in Story Builder and Deposition Analyzer would be included in the core per-gigabyte rate, that it was "not increasing its per-gigabyte pricing," that batch coding suggestions would be priced more than 40% lower, and that Deep Dive would carry "a one-time per-gigabyte ingestion fee" with unlimited questions for the life of a case.
The live pricing page still marks Coding Suggestions, Deep Dive, Machine Translation and Custom Extractions with "credits"; all but Deep Dive note that single-document use needs none. It lists unlimited user licenses and no dollar figures.
The only public Everlaw dollar figures are on its GSA federal price list (August 2025 file, FedRAMP platform, net federal prices). Active data on a per-case subscription is $25.24/GB/month; bundle overage rates fall from $17.42/GB at 500 GB to $7.49/GB at 50 TB. Suspended data is $3.50/GB/month. AI Assistant credits cost $362.72 a month for up to 10,000 credits, $0.0363 per overage credit, or $0.0453 per credit pay as you go. The list says credit use "will vary depending on the complexity and number of pages": a writing-assistant task "may cost 300 credits," a 10-page summary "just 1 credit." It also notes subscriptions include unlimited processing, productions and exports, and that suspended data is hidden from all users until reactivated.
Two cautions. The federal list predates the October 2025 changes, so commercial batch AI should be cheaper than it implies. And Everlaw publishes no credits-per-document figure for batch coding suggestions, so you cannot price an AI-led review of a large set without a quote.
Who should not buy Everlaw: a team that plans to run generative first-pass review across hundreds of thousands of documents per matter and wants that cost fixed. The metered credits are the variable you will be negotiating every time.
What DISCO's All-Inclusive Price Covers
DISCO's per-GB fee covers hosting and the Cecilia assistant features, and leaves out Auto Review, which is the product that makes responsiveness calls. Its eDiscovery pricing page says the fee, based on processed data size, covers hosting in Active Review and ECA plus Cecilia Q&A, document summaries, Auto Timelines and Deposition Summaries, with "no separate fees for ingest, processing, or users" and no charge for duplicates removed at ingest. It publishes no rate and no minimum.
The Auto Review pricing page lists a "per document charge" and asks you to contact sales. It claims review speeds of "more than 10k docs per hour" in one place and "up to 32k docs per hour" in another.
On DISCO's Q1 2026 earnings call, CEO Eric Friedrichsen said the "old pricing model was hard to understand" and made DISCO look "much more expensive than the competition especially for larger matters," and that the new model was "designed to increase consideration, to improve win rates, to reduce discounts." Management also said Auto Review revenue is split between software and services, with manual prompt engineering still sitting in services. That is a signal to ask whether your Auto Review quote includes the prompt work or bills it as hours.
DISCO also sells DISCO Advanced Research; this comparison covers only its eDiscovery pricing.
Why DISCO comes last for this buyer: the headline promise is "unlimited AI," but the AI that replaces first-pass reviewers is priced per document, off-page, and partly as services. Relativity includes its equivalent in committed volume; Everlaw at least publishes the unit (credits). DISCO also publishes no vault or archive rate on its pricing pages.
Who should not buy DISCO: anyone who must give a client or CFO a fixed AI review budget before collection finishes.
A Worked Cost for One 500,000-Document Matter
Generative AI review cuts this matter's review bill by roughly 40% to 90%, and how the AI line is contracted moves the total far more than the per-GB rate does. The workload, stated so you can swap in your own numbers: 500,000 documents promoted to review, 100 GB hosted in review for 6 months, then 18 months dormant before the matter closes. 10% of the set (50,000 documents: privilege candidates, QC samples and AI exceptions) goes to human reviewers.
| Line | Linear human review | GenAI review, survey pricing | Relativity, committed | Everlaw (GSA list) |
|---|---|---|---|---|
| First-pass review | $250,000 to $500,000 (500k × $0.50 to $1.00) | $55,000 to $250,000 (500k × $0.11 to $0.50) | $0 incremental for aiR | Credits; at an assumed 1 credit/doc, $22,650 |
| Human QC and privilege (50k docs × $0.50 to $1.00) | Included above | $25,000 to $50,000 | $25,000 to $50,000 | $25,000 to $50,000 |
| Review hosting, 6 months, 100 GB | Varies | Varies | £4,200 (Epiq card, £7.00) | $15,144 ($25.24) |
| Dormant storage, 18 months, 100 GB | Varies | Varies | £2,700 (Epiq card, £1.50) | $6,300 ($3.50) |
The one-credit-per-document figure is an assumption, used only to show scale: Everlaw does not publish the batch coding-suggestion rate, and the in-platform estimate is what you should get in writing. Processing is included on Everlaw's list and priced at £27.65 to £45.50/GB on Epiq's. DISCO is absent from the table because it publishes neither number; plug its quote into the survey column.
Three things fall out of the arithmetic. Hosting and storage together are under $25,000 in every column, so a per-GB discount is worth less than one point off a per-document AI rate. The AI line ranges from $0 to $250,000 depending on contract structure alone. And the human line never disappears: the most defensible workflows still put tens of thousands of documents in front of lawyers.
What Courts and Opposing Counsel Have Accepted
Courts are treating generative AI review like technology-assisted review (TAR): permitted, with disclosure governed by your ESI protocol, and validation metrics compellable only on a showing of a specific deficiency. TAR here means any process where software, rather than a lawyer, decides which documents are responsive.
In Schulte v. LinkedIn (N.D. Cal., June 30, 2026), Magistrate Judge Laurel Beeler let LinkedIn pre-cull with 25 search strings before running Relativity aiR and refused to compel elusion estimates or reviewer counts, because "discovery on discovery is disfavored absent a showing of a specific production deficiency." LinkedIn's disclosures were modest: no seed or training set, aiR making final responsiveness calls, and human review of samples from each category.
Negotiated protocols can demand far more. The ESI protocol in James v. Cerebras Systems (N.D. Cal., Magistrate Judge Robert Illman, July 7, 2026) sets an expectation that elusion "won't exceed 3%," requires validation sampling at 95% confidence with a 2% margin of error, and requires disclosure of all prompts, with any prompt change sent in redline within three business days. A 95%/±2% sample of the excluded set is roughly 2,400 documents, which is a budget line you should add before agreeing to it.
Privilege review carries the largest savings and the largest waiver risk. An Arnold & Porter attorney who used aiR for Privilege on an antitrust review reported about 75% savings against an estimated contract-attorney linear review. The tool favoured recall, so it over-flagged and created "in between" categories that needed human review, and classifying each entity's alignment with the client took one attorney close to two weeks on about 500,000 documents. At a Legalweek 2026 panel, Morgan Lewis's Liz Gary said: "I don't know if I would necessarily opt for an AI tool without that protection of a 502 order." Get a Federal Rule of Evidence 502(d) clawback order entered before any AI privilege pass.
The Storage Fees That Keep Running After a Matter Goes Quiet
Every platform keeps billing for a stayed or settled matter until someone moves it to a cheaper tier, and the move itself has rules. Relativity's data management documentation says a workspace moved to cold storage is "billed at their original data tier rate until the start of the following calendar month," and a review workspace converted to repository stays at the review rate until the next month too. Moving data on the 2nd costs you a full month at the higher rate.
Partner contracts add reactivation labour. Epiq's list says reactivation of dormant workspaces is quoted per workspace and billed at technical analyst rates (£145/hour on the same card). Everlaw's suspended tier is cheap on its federal list but hides data from all users, so suspend only what nobody will need on short notice. DISCO's pricing pages do not state a vault rate at all, so get it in the order form before you sign.
How to Choose: The Criteria That Predict Regret
Rank vendors on AI metering, exceptions, validation and exit terms before the per-GB rate, because the worked example shows the per-GB rate moves the total least. In order:
- Matter volume per year. Three or more review matters: commit to Relativity and take the included aiR. One or two: Everlaw's per-case subscription is easier to live with.
- How AI usage is counted: per document reviewed, per document ingested (Epiq's model), per credit, or included. Ask for the count on a sample set before signing.
- Exception handling: which documents the AI fails on, and at what rate they are reviewed. The survey shows many buyers do not know.
- Validation cost: whether the vendor or partner includes elusion sampling and the prompt-disclosure work a protocol like Cerebras demands.
- Exit and dormancy: cold, suspended or vault rates, the billing cut-over date, reactivation fees and export costs.
What changes the answer: if Relativity starts metering aiR under a "fair use" cap, Everlaw's single-document inclusion looks better. If DISCO publishes Auto Review pricing at the low end of the survey band, it becomes a reasonable choice for teams that want one vendor for review and deposition work.
This Week:
- Pull your current eDiscovery contracts and find the AI line. If it is per document on Relativity, ask to move to a committed plan.
- Ask every vendor in writing how many documents in your last matter would have been AI exceptions, and what rate they would bill.
This Month:
- Run a priced pilot on a closed matter: same 50,000 documents through each platform's AI, with elusion sampling, and compare the invoices line by line.
- Get a 502(d) order template approved by your litigation group for any matter where AI touches privilege.
Before Your Next Renewal:
- Negotiate dormant and cold storage rates, the tier cut-over date and reactivation fees into the order form.
- Write the validation sampling standard (confidence level, margin, who pays) into your ESI protocol template.
The Bottom Line
eDiscovery pricing is following the path TAR took: once the software became cheap, negotiations moved to validation and to what counts as a billable unit. In 2026 the AI review line ranges from zero to a quarter of a million dollars on the same 500,000 documents, depending only on contract structure. Relativity's committed-volume inclusion is the lowest marginal AI cost of the three for repeat buyers; Everlaw is the predictable choice for small matters; DISCO has to publish its Auto Review price before it can be evaluated on the same footing.
Ask each vendor to price your last matter, AI exceptions and storage included, before you compare a single per-GB rate.
Continue Reading
- Harvey vs CoCounsel vs Legora: Buy the Library, Not the Model
- Harvey Swapped Models After Agents Sank Its Margin to -50%
- CoCounsel's New Model Runs on Qwen. Go Read the Card.
- Astra for Law Scored 54% on a Set Vals Never Publishes
- Clio's Learned Hand Deal Puts Lawyers and Judges on One AI Vendor
- Inference Cost per Million Tokens: 2026 Provider Pricing Compared