Mercor pricing

About

Mercor is an AI talent and data marketplace that connects vetted human experts — doctors, lawyers, engineers, financial analysts, consultants — to AI training, evaluation, and agent-building projects run by frontier AI labs and enterprises. The thesis is that frontier models no longer improve on public data alone; they improve when domain experts apply professional judgment to how models reason and respond. Mercor packages that expertise as placeable, scoped contract work and, increasingly, as enterprise data partnerships and managed agent deployments.

The company operates three connected businesses: a talent marketplace (Experts) that finds, vets, and places professionals into AI projects “in days,” with 4M+ experts in its network; an enterprise data-partnership business — rebranded across the site in July 2026 as Human data and Data monetization — that pays companies to anonymize and license their operational workflow data to AI labs across 50+ tool integrations; and an Enterprise agents practice — reframed 2026-08-11 as Discover → Deploy → Improve → Monetize (agent diagnostics, deployment, optimization, and data monetization; “optimization” replaces the prior “benchmarking” label) — promising production agents in 4–6 weeks. A separate APEX research line (APEX Benchmarks, APEX-Agents, APEX-Accounting, APEX-SWE, Off-the-shelf data) sits alongside them in the site nav — APEX-Accounting is a new addition first observed 2026-08-04.

Through 2026-08-04, Mercor’s Enterprise page carried a headline scale bar — $10B valuation, $2B+ revenue run rate (up from the $1B+ it advertised through June 2026), 100k+ contractors hired, and 400+ full-time employees — plus an “Our first customer was ourselves” case study (90% of support resolved end-to-end by AI, 12k+ requests resolved, $1.4M saved by support automation, a 73% reduction in ticket close time, 70% of support tickets handled or assisted by AI). As of 2026-08-11 the Enterprise page was rebuilt and no longer shows either — the scale-stats bar and the self-referential case study are both gone, replaced by three named client engagements and a new Discover → Deploy → Improve → Monetize framing of the four offerings (see Pricing by product). As of 2026-08-12 that case-study set grew to six, adding a Workflow Diagnostic (25 candidate AI workflows identified across seven functions), a second Value Creation Assessment for a PE firm ($20M in annual run-rate impact identified by 2031, plus ~$4M in near-term net value), and a Production Readiness Evaluation (132 expert-graded agent runs across real-world HR tasks). The $10B valuation and October 2025 Series C terms remain independently documented by press (TechCrunch/CNBC, 2025-10-27). Mercor is headquartered in San Francisco and remains privately held.

Pricing summary : marketplace take-rate, with only talent pay disclosed

Mercor runs a sales-gated marketplace in which the only publicly visible prices are talent-side pay rates — never the buyer-side price. The relevant dimensions:

1. Talent rate (visible, contractor-side): Mercor’s Experts page advertises an “average contracted rate” of $112/hr (as of 2026-08-26, up from the previous capture — the first weekly uptick after several weeks of drift lower; this headline metric has been volatile since its June 2026 relabel — see Pricing evolution for the full swing), alongside 407.9K roles created (up from 401.9K) and $4M+ in daily payouts. Individual roles are posted between roughly $60/hr and $250/hr depending on domain and seniority (e.g. a “Cybersecurity Research Expert — Offensive Security & Vulnerability Research” Talent Network role at $200–$250/hr, “Law Experts” at $110–$150/hr, Physician Talent Network $110–$250/hr, Machine Learning Engineer Talent Network $70–$250/hr, a “Legacy Codebase Migration Expert” at $200/hr). Experts are paid weekly. This is what Mercor pays contractors — it is not what buyers pay Mercor.
2. Buyer price (undisclosed, sales-only): What AI labs and enterprises pay Mercor — the marketplace take-rate / margin on placed expert hours, plus engagement fees for agent diagnostics, deployment, and optimization — is never published. Every buyer surface terminates in a “Talk to the team” contact form or “Book a demo” CTA. Third-party indicative only: analyst firm Sacra (via secondary coverage at eesel) pegs the employer-paid recruiting fee near ~30% of candidate pay and describes hourly work as cost-plus — useful as a rough anchor, but not a Mercor-confirmed rate. As of 2026-08-11, the Enterprise page was rebuilt: its four offerings are now framed as Discover → Deploy → Improve → Monetize (the “Improve” step — formerly “Mercor Agent Benchmarking” — is now “Mercor Agent Optimization,” reviewing agents already in production rather than benchmarking a client’s own AI products), proof was now three named case studies (a PE-firm pre-deal Value Creation Assessment identifying $7–9M upside run-rate savings; a technology SRE engagement citing a 50%+ reduction in median time to mitigation; Mercor’s own customer-support deployment now cited at ~12,000 support requests handled per week), and the prior scale-stats bar (valuation, revenue run rate, contractor count, employee count — see About) and “Our first customer was ourselves” case study no longer appear on the page. As of 2026-08-12 that proof set expanded to six named case studies, adding a Workflow Diagnostic (25 candidate AI workflows identified across seven functions), a second PE-firm Value Creation Assessment ($20M in annual run-rate impact identified by 2031, plus ~$4M in near-term net value), and a Production Readiness Evaluation (132 expert-graded agent runs across real-world HR tasks) — still with no price attached to any engagement. As of 2026-08-25 the same four offerings and six case studies hold over, but each offering step’s illustrative product-screen mockup was refreshed (a workflow-diagnostic dashboard now illustrates Discover, an invoice-reconciliation run summary illustrates Deploy, an effectiveness-vs-cost model chart illustrates Improve, and a partially-populated tool-selection/licensing-opportunity panel illustrates Monetize) and several “what you get” bullets under Deploy and Monetize were reworded — still with no price attached to any engagement.
3. Data-partnership compensation (contributor-side): On the Data page — now surfaced in the nav as Human data and Data monetization — Mercor pays enterprises to license anonymized operational data, scaling compensation with data volume, data depth, and the number of tools connected (50+ integrations as of July 2026, up from 34+; “connecting more tools generally results in a higher payout”), with “no fees or costs” to the contributor and payment by wire within 2–4 weeks of a signed agreement. As of 2026-07-29 the Data page also advertises a company-referral incentive — “Earn up to $100K per referral” — the first time this figure has been disclosed; the same slot previously read only “if it works out, we’ll pay you a referral bonus” with no amount shown.

What makes this different: Mercor inverts the usual pricing-page convention — it loudly publishes what it pays out (hourly expert wages, data-partner payouts) while keeping what it charges entirely behind sales, so the headline dollar figures on the site are payouts, not the product price. This is a classic usage-based pricing marketplace where the metered unit (expert-hours / scoped projects) is real but the buyer rate is opaque, and the whole motion is sales-led.

Pricing by product

Important: Every dollar figure below is a talent-side pay rate or a data-contributor payout — money Mercor pays out. It is not the buyer-side price. What AI labs and enterprises pay Mercor (the marketplace take-rate / engagement fee) is undisclosed and sales-quoted on every surface.

Buyer-side engagements (price undisclosed)

Third-party indicative only (not Mercor-confirmed): Sacra’s analyst note, via eesel’s pricing teardown, estimates the employer recruiting fee near ~30% of candidate pay and characterizes hourly expert engagements as cost-plus. Treat these as rough anchors — Mercor publishes no rate card, so the realized margin per engagement is not verifiable from any first-party source.

Talent rates (contractor-side, public — NOT the buyer price)

The “Latest roles” board rotates; the table above is the live listing set captured on 2026-08-26. Three listings from the 2026-08-25 capture rotated off (Architecture Expert, STEM Expert (Physics, Chemistry, Mathematics, Biology), and the Employer Benefits Decision-Makers Survey), replaced by a new flat-rate “FP&A Expert” listing at $2K and the reappearance of two engineering roles (ML Engineer and Data Engineer, both “Coding Agent Experience”) that had rotated off on 2026-08-25 after last appearing 2026-08-12. Individual role rates turn over faster than the headline average. The overall published band has stayed $60–$250/hr throughout.

Data-partnership compensation (contributor-side, paid TO the enterprise)

Sales motions across products: there is no PLG / self-serve tier anywhere — talent placement, enterprise agents, and data partnerships are all sales-led and engagement-scoped, with buyer pricing custom-quoted via “Talk to the team” forms.

Hidden costs : the take-rate is the hidden cost

There is no public Mercor bill to itemize, so a synthetic cost table would be fabrication. The honest framing is the opposite of most pages in this corpus: the hidden cost is the undisclosed take-rate itself. The only dollar figures Mercor publishes are contractor rates — what it pays experts ($60–$250/hr, ~$121/hr advertised average “contracted rate” as of 2026-07-21) and data partners (custom). What a buyer pays for those same hours is never shown, so buyers cannot reconcile their invoice against the talent-side rate they can see on the public marketplace.

What is known, and where the cost hides:

• The markup is invisible by design. A buyer hiring an expert posted at $60–$250/hr does not know whether Mercor’s fee is 10%, 30%, or 50% on top. Third-party analysts (Sacra via eesel) estimate a ~30% recruiting fee and cost-plus hourly billing, but Mercor confirms nothing — so the effective blended rate a lab pays is unverifiable.
• Engagement scope is the real meter. Enterprise services (Agent Diagnostics, Deployment, Benchmarking, Data Monetization) are quoted per engagement, not per seat or per token, so two buyers with similar headline needs can pay very different totals depending on how the SOW is scoped.
• Data-partnership “free” has a catch worth naming. Mercor tells data contributors there are “no fees or costs,” and that is true for the contributor — but it is silent on what the AI-lab buyer pays for that same anonymized dataset, which is the margin that funds the contributor payout.
• Procurement risk is a cost too. Because there is no rate card, every deal is a custom negotiation; without a reference price or a clear value metric, buyers carry the full information asymmetry, which the sales-led motion is built to exploit.

Illustrative buyer cost (third-party indicative only — NOT a Mercor quote)

There is no Mercor rate card, so the table below is not a real bill. It simply applies the only two numbers available — the public talent rate and a third-party take-rate estimate — to show the shape of what a buyer might pay for one expert, full-time-equivalent, for a month. Every figure except the talent rate is an outside estimate, not a Mercor-confirmed price.

The lesson is not the number — it is that a buyer cannot derive this from any Mercor surface. The $121/hr is real and public; the ~30% markup is an outside estimate; the total is therefore a guess, which is exactly the procurement problem Mercor’s opacity creates.

Want to model your own scenario? Use the Mercor pricing calculator to sketch buyer cost across expert rate and an assumed take-rate — bearing in mind that, unlike most pages in this corpus, every output is an indicative estimate built on third-party assumptions, not a Mercor quote. To compare opaque, sales-quoted usage-based pricing motions against transparent ones, browse the pricing blueprint corpus.

Pricing evolution : from talent marketplace to enterprise data + agents

Mercor’s “pricing” evolution is really a business-model evolution: the buyer-side price has been sales-gated from day one, so what changes over time is the shape of what’s being sold (AI-interview recruiting → human-data/RLHF labeling → enterprise agents + data partnerships) and the talent-side payouts that are publicly advertised. The cadence below tracks model and payout shifts, not list-price changes — because there is no public list price.

Cadence

Tracked range: 2025-02–2026-08. Wayback archives begin 2026-03 for the Enterprise/Experts surfaces; pre-2026 model history is reconstructed from press. No public buyer price has ever been listed, so the “price changes” column tracks advertised talent payouts, not buyer list prices.

Notable changes

• 2025-02-20 — Series A: $100M at $2B valuation; AI-interview recruiting marketplace, ~$85/hr avg pay, ~30k contractors (TechCrunch).
• 2025-10-27 — Series C: $350M at $10B valuation, Felicis-led with Benchmark, General Catalyst, Robinhood Ventures (TechCrunch).
• 2026-03 — Enterprise positioned as an agent-building software platform (Wayback 2026-03 Enterprise snapshot); average advertised pay ~$99/hr.
• 2026-03/04 — LiteLLM supply-chain breach exposed Slack, ticketing, and contractor-interaction data; Meta paused work (TechCrunch, Wired).
• 2026-05 — Standalone Data-partnership business surfaces (34+ OAuth integrations, SOC 2 Type II, 2–4 week payouts); average advertised pay rises to $105/hr (Wayback 2026-05).
• 2026-06-07 — Enterprise restructured into four named, sales-quoted services (Agent Diagnostics, Deployment, Benchmarking, Data Monetization); average advertised pay reaches $141/hr, 258.1K roles created, $3M+ daily payouts (Wayback 2026-06 + live capture).
• 2026-06-30 — The Experts hero metric is relabeled from “average pay” to “average contracted rate” and the figure drops from $141/hr to $80/hr; roles created rise to 287.1K, daily payouts unchanged at $3M+ (live capture).
• 2026-07-09 — Mercor acquired Deeptune (Andreessen Horowitz-backed) to build training environments and simulators for AI agents — its first expansion beyond human-talent matching into the RL-environment supply chain that feeds the same frontier-lab demand its expert hours serve (Fortune / SiliconANGLE / The Information, 2026-07-09). Deal terms were not disclosed and no buyer pricing changed; this is an ownership / supply-side move, not a rate move.
• 2026-07-14 — Two weeks after the reset, the “average contracted rate” rebounds from $80/hr to $122/hr (a ~52% swing back up); roles created rise to 305.0K and daily payouts to $4M+, while the Enterprise revenue run rate doubles from $1B+ to $2B+ (live capture).
• 2026-07-21 — The buyer-facing product taxonomy is renamed in the nav and footer of all four surfaces: the “Business” column becomes “Enterprise” and its rows go Enterprise → Enterprise agents, Enterprise evals → Human data, Data partnerships → Data monetization, with the APEX research line (Benchmarks, Agents, SWE, Off-the-shelf data) sitting alongside. In the same capture the Data page’s payout base widens from 34+ to 50+ integrations, and the headline “average contracted rate” holds essentially flat at $121/hr (from $122/hr) on 313.9K roles created and $4M+ daily payouts (live capture).
• 2026-07-29 — The Data page discloses its first concrete referral-bonus figure — “Earn up to $100K per referral” — for enterprises referred into the data-partnership program, replacing a prior undisclosed-amount promise (“if it works out, we’ll pay you a referral bonus”) and adding a new expandable “How it works” 3-step flow (refer a company → they onboard → you get paid once their payout settles). Talent-side numbers moved only within routine weekly noise: the “average contracted rate” ticked from $121/hr to $120/hr on roles created rising from 313.9K to 323.7K; the “Latest roles” board rotated its usual weekly set within the existing $60–$250/hr band, and enterprise scale stats ($10B valuation, $2B+ run rate, 100k+ contractors, 400+ employees) and the 50+ integrations figure were unchanged (live capture).
• 2026-08-11 — Mercor rebuilt its Enterprise page: the prior scale-stats bar ($10B valuation, $2B+ run rate, 100k+ contractors, 400+ employees) and the “Our first customer was ourselves” case study were removed; “Agent Benchmarking” was renamed “Agent Optimization”; proof shifted to three named client engagements (a PE-firm pre-deal Value Creation Assessment citing $7–9M in identified savings, an SRE engagement citing a 50%+ reduction in time-to-mitigation, and Mercor’s own support operation cited at ~12,000 requests/week). Talent-side headline eased from $120/hr to $118/hr on roles created rising from 330.6K to 339.5K (live capture).
• 2026-08-12 — One day later, the Enterprise page’s case-study count doubled from three to six: added a Workflow Diagnostic (25 candidate AI workflows identified across seven functions), a second PE-firm Value Creation Assessment ($20M in annual run-rate impact identified by 2031, plus ~$4M near-term), and a Production Readiness Evaluation (132 expert-graded HR-agent runs); two of the original three case studies were also relabeled at the category level, and a newsletter signup block was added near the page footer. Buyer pricing remained undisclosed throughout. Talent-side headline eased from $118/hr to $114/hr — its steepest single-day drop since the June reset — on roles created rising from 339.5K to 349.3K (live capture).
• 2026-08-25 — The Experts hero “average contracted rate” eased further to $114/hr → $109/hr on roles created rising from 349.3K to 401.9K, its lowest reading since the mid-July rebound; the live “Latest roles” board also showed heavier-than-usual turnover (seven listings rotated off, seven new ones appeared, including a $200/hr “Legacy Codebase Migration Expert” and two flat-rate survey gigs at $70 and $63). Separately, the Enterprise page’s four offering steps (Discover/Deploy/Improve/Monetize) kept their same names and the same six named case studies, but each step’s illustrative product-screen mockup was replaced — a workflow-diagnostic dashboard now illustrates Discover (previously a voice-AI interview demo), an invoice-reconciliation run summary illustrates Deploy (previously an agent-configuration screen), an effectiveness-vs-cost model-comparison chart illustrates Improve (previously a verifier-score table), and a partially-populated tool-selection / data-licensing-opportunity panel illustrates Monetize (previously a blank tool selector) — and several “what you get” bullets under Deploy and Monetize were reworded. Buyer pricing remained fully undisclosed throughout (live capture).

The July 2026 rename in detail

The rename is small in words and large in what it reveals. Every old label named either a segment (“Business”) or a service (“Enterprise evals”, “Data partnerships”); every new label names a thing being bought — Enterprise agents, Human data, Data monetization. That is the vocabulary shift a company makes when it stops selling projects and starts selling products, and it is normally the step immediately before a price appears. Here it didn’t: all three lines still terminate in the same “Talk to the team” form, so Mercor has acquired product nouns without acquiring product prices.

Two of the substitutions carry real signal. “Enterprise evals” → “Human data” collapses an evaluation service into a data commodity, which is consistent with the underlying reality that the eval practice and the data-partnership practice draw on one supply base — expert hours — and differ mainly in how the output is packaged for the buyer. “Data partnerships” → “Data monetization” flips the frame from Mercor’s relationship to the contributor’s revenue line, which is a sharper pitch to a CFO but also quietly names the spread: the enterprise “monetizes” its data, Mercor sells that same data to labs, and only one of those two prices is ever shown.

The integration count is the other half of the change, and it is closer to a pricing move than a feature move. Mercor tells contributors that “connecting more tools generally results in a higher payout,” which makes the number of supported integrations the closest thing to a published meter anywhere on the site. Going 34+ → 50+ widens that meter’s base by roughly half — more connectable surface per contributor, so a bigger dataset and a bigger payout per deal — without disclosing the per-tool rate. The countable unit got bigger; the price per unit stayed invisible, so a contributor’s expected payout became larger and less estimable at the same time.

In late March 2026, Mercor was caught in a supply-chain attack on the open-source LiteLLM project: a hacking group (TeamPCP) inserted malicious code that was removed within hours, after which the extortion group Lapsus$ claimed it had targeted Mercor’s systems. The exposed material reportedly included Slack data, ticketing information, and videos of AI-system-to-contractor conversations; Mercor declined to confirm whether customer or contractor data was exfiltrated, characterizing itself as “one of thousands of companies” affected. Wired reported that Meta paused work with Mercor in response, and the incident drew a 600-point, 226-comment Hacker News thread (2026-04-27) — a notable trust event for a marketplace whose entire buyer relationship is opaque and sales-gated. For a price-secret vendor, a breach that touches contractor data is precisely the kind of event that erodes the asymmetric trust the model depends on.

What’s unique : publishing payouts while hiding the buyer price

1. Payouts are public; the price is secret — and so, increasingly, is the shape of the proof. Almost every company in the pricing blueprint publishes what it charges and hides its costs. Mercor inverts this: it loudly advertises what it pays out — a $109/hr advertised average “contracted rate” (2026-08-25, down from $114/hr two weeks earlier and its lowest reading since the mid-July rebound), $60–$250/hr by role, $4M+ in daily payouts, a 50+ tool payout base on the data side, and — as of 2026-07-29 — a first-ever disclosed referral-bonus figure of up to $100K per referral for the data-partnership program — while keeping the buyer-side take-rate entirely behind a contact form. The same asymmetry now shows up on the Enterprise page’s proof section: it doubled from three to six named case studies in a single day (2026-08-11 → 2026-08-12), each anchored to a concrete quantified outcome ($7–9M in identified savings, a 50%+ reduction in time-to-mitigation, 25 candidate workflows identified, $20M in projected annual run-rate impact by 2031, ~12,000 support requests handled weekly, 132 expert-graded agent runs) — yet none of the six carries a dollar figure for what the engagement itself cost. The headline dollar figures on the site are wages and client outcomes, not the product price, which is a deliberate recruiting-and-credibility funnel design that also conveniently obscures Mercor’s margin.

2. The “AI interview” is the vetting mechanism and the moat. Mercor screens experts through an adaptive AI interview that “evaluates expertise at scale,” asking field-specific questions to assess depth. This lets a tiny team vet a 4M+ expert network without human recruiters — it is the operational lever behind the marketplace’s speed (“first offer instantly,” placement “in days”) and a genuine differentiator versus traditional staffing firms or labeling vendors like Scale AI / Surge.

3. A marketplace take-rate dressed as enterprise infrastructure. Underneath the agent-deployment and data-partnership packaging, the core economic engine is a classic two-sided marketplace take-rate: Mercor pays the supply side (experts, data contributors) and charges the demand side (AI labs, enterprises) a margin it never discloses. The enterprise-services and data-monetization SKUs are higher-margin wrappers on the same human-expertise supply.

4. It pays the supply side on both sides of its data business — and since July 2026 the nav names both sides separately. In the data-partnership business, Mercor pays enterprises to license their anonymized operational data, then sells access to AI labs — so it is buying from one set of customers and selling to another, monetizing the spread. The 2026-07-21 rename made that structure legible: the single “Data partnerships” entry split into Human data (what the lab buys) and Data monetization (what the contributor earns), so the two ends of the spread now have separate names in the same menu. Naming both ends without pricing either is the sharpest expression of Mercor’s model — the contributor still sees “no fees,” which is true, and still never sees the price the lab pays for their data.

Strengths & weaknesses

Billing UX : contact forms, weekly payouts, and OAuth-scoped data

• “Talk to the team” / “Talk to the Team” contact form (Experts/Partner) — the universal buyer entry point: First name, Last name, Business email, Company, Job title, “What can we help with?” dropdown, plus a free-text field. There is no checkout or rate card.
• “Book a demo” / “Talk to the Mercor team” CTAs (Enterprise, rebuilt 2026-08-11; offering-step demos refreshed by 2026-08-25) — every enterprise offering, now framed as Discover → Deploy → Improve → Monetize (Agent Diagnostics, Agent Deployment, Agent Optimization — renamed from Agent Benchmarking, and Data Monetization), routes to scheduling time with sales rather than a price. The page embeds an interactive product-screen preview for each step — a workflow-discovery dashboard (org chart, per-function AI-potential scoring) under Discover, a live invoice-reconciliation run summary under Deploy, an effectiveness-vs-cost model-comparison chart under Improve, and a partially-populated tool-selection / data-licensing-opportunity panel under Monetize — but none of these surfaces a price.
• Weekly contractor payouts — Mercor advertises “Daily payouts $4M+” in aggregate alongside “Roles created 407.9K” in the Experts hero bar (up from 401.9K on 2026-08-25), and tells experts they “get paid weekly for contributing your expertise”; payment terms are “defined upfront” per project. Its talent help center pins the cadence down: contractors are paid every Wednesday around 12:00 PM PST for the prior Saturday 12:00 AM → Friday 11:59 PM IST work week (~five-day lag), with bonuses, work-trial pay, and referrals on a separate two-window-per-week (Wednesday/Friday PST) schedule. All contracts are denominated in USD, converted to local currency at payout.
• Payout speed is the one percentage fee Mercor publishes — contractors choose between Standard (free, funds arrive “up to 5 business days” after release) and Instant (a 1.0% fee, ~30 minutes, where the bank supports Stripe Instant Payouts), switchable under Profile → Account → Payout preferences. Payouts run through Stripe Connect, or Wise in countries Stripe Connect does not cover (Wise requires the expert to action an email before funds move). This 1.0% is a talent-side convenience charge, not the buyer-side take-rate — which remains undisclosed.
• Data-partner OAuth onboarding — contributors “authenticate via OAuth” to grant read-only access across 50+ integrations (raised from 34+ in July 2026) spanning messaging, docs, email, CRM, code, and finance, then receive a single background extraction; payout is wired within 2–4 weeks of delivery, followed by a custom AI-transformation analysis.
• Data scoping + referral forms as the only “checkout” — the Data page offers a “Start here →” scoping form (“share what services you’re ready to license”) and a separate referral module now headlined “Earn up to $100K per referral” (disclosed 2026-07-29; previously an undisclosed “if it works out, we’ll pay you a referral bonus”), with an expandable “How it works ▾” 3-step flow (refer a company → they onboard → you get paid once their payout settles) plus the same referral form fields (name, email, referee, company website, employee-count band 1-50 / 51-200 / 201-1,000 / 1,000+). The underlying data-licensing scoping form still shows no price.
• Talent-network “Apply once” model — experts apply once to a Talent Network listing (e.g. Physician $110–$250/hr) and receive future offers that match their qualifications, with referral links and referral bonuses surfaced on each job page.

Strategic wins : why the sales-gated marketplace motion worked

1. Advertise payouts, not prices — turn the recruiting funnel into the marketing engine

By making the talent-side pay the loudest number on the site, Mercor solves its hardest problem (sourcing scarce domain experts) while keeping its margin invisible. The public headline figure and $4M+ daily payouts are simultaneously a recruiting pitch and a credibility signal to buyers (“look how serious our supply is”). That it could quietly relabel the metric from “average pay” to “average contracted rate” and reset it from $141/hr to $80/hr on 2026-06-30, then let it rebound to $122/hr by 2026-07-14 — all with no buyer-facing explanation — only underscores how fully Mercor controls the one number it does publish. It is a textbook example of choosing which side of a two-sided market to make transparent — and the value-metric problem resolves in Mercor’s favor because the metric it shows (wages) is not the metric it bills on.

2. Use sales-gating to capture maximum margin in a nascent, undifferentiated market

With no public rate card, every engagement is bespoke and the buyer carries the information asymmetry — the defining feature of a sales-led pricing motion. In a market where comparable RLHF/data vendors (Scale AI, Surge) are also opaque, transparency would only arm procurement; opacity lets Mercor price each frontier lab to its willingness to pay. This is the right call while the category is young and reference prices don’t exist.

3. Build adjacent, higher-margin SKUs on the same human-expertise supply

Mercor turned one supply base (vetted experts) into three demand-side products: talent placement, enterprise agents, and data partnerships. Each new SKU monetizes the same network at a higher margin without re-acquiring supply — a capital-efficient expansion that helped drive the move from a $2B to a $10B valuation in eight months. The 2026-07-21 rename finished the job at the label layer: “Business”, “Enterprise evals”, and “Data partnerships” became Enterprise agents, Human data, and Data monetization — segment and service words replaced by the names of purchasable things. Renaming an eval service into a data commodity (“Enterprise evals” → “Human data”) is the tell that the underlying asset was always the same expert supply, packaged differently per buyer. It mirrors how successful usage-based pricing businesses layer outcome-based wrappers on a metered base. The one departure from the same-supply logic arrived on 2026-07-09, when Mercor acquired Deeptune to build AI-agent training environments — its first move to add a new supply category (RL environments / simulators) rather than repackage the existing expert network. It is aimed at the same frontier-lab demand, but whether that new supply carries the same undisclosed take-rate economics, or a different pricing shape, is not yet evidenced.

4. Pay the supply side to unlock a brand-new data asset

The data-partnership model pays enterprises for something they previously gave away or sat on — their operational workflow data — and resells it to labs. By absorbing all extraction/anonymization cost and charging the contributor nothing, Mercor removes the friction that kills most data-licensing deals, then monetizes the spread on the buyer side. Widening the connector base from 34+ to 50+ tools on 2026-07-21 is a payout-side lever disguised as a feature release: because payout scales with tools connected, every added integration raises both the dataset Mercor can resell and the number the contributor sees on the wire, so supply-side acquisition and cost of goods move together in the right direction. It is a clean example of aligning the meter to value created rather than to cost incurred.

5. Replace vanity scale-stats with quantified, named case studies as the credibility mechanism

When Mercor rebuilt its Enterprise page on 2026-08-11, it swapped a broad scale-stats bar ($10B valuation, $2B+ run rate, 100k+ contractors, 400+ employees) and one self-referential case study for three named client engagements, each anchored to a specific number ($7–9M in identified savings, a 50%+ reduction in time-to-mitigation, ~12,000 requests/week). One day later, on 2026-08-12, it doubled that set to six, adding a Workflow Diagnostic (25 candidate workflows), a second PE-firm assessment ($20M in projected annual run-rate impact by 2031), and a Production Readiness Evaluation (132 expert-graded agent runs). For a marketplace that cannot show a price, this is the next-best trust signal: instead of asking a buyer to believe an aggregate valuation number, it asks them to believe a specific, attributable client outcome. Quantified proof lets a buyer benchmark expected value even without a benchmark on cost — but it only works as long as each case study stays this specific; six generic client logos would not do the same work as six dollar-and-percentage outcomes.

Areas to improve : buyer-side price opacity and trust gaps

1. Publish at least a pricing framework, even without a rate card

Total opacity maximizes margin but also maximizes buyer friction and suspicion — the recurring “the final price tag for the client remains a mystery” critique in third-party teardowns. Mercor could publish a structure (e.g. “cost-plus on placed hours; engagement-scoped for enterprise services”) without revealing the actual percentage, the way many enterprise vendors disclose their pricing model while keeping the number gated. That alone would reduce procurement anxiety without surrendering negotiating leverage.

2. Close the trust gap the LiteLLM breach opened

A breach that touched contractor data — and made a major lab pause work — is existential for a marketplace built on opaque, asymmetric trust. Mercor’s terse, non-committal incident response (declining to confirm what was exfiltrated) is the opposite of what rebuilds confidence. Publishing a concrete post-incident security posture (scope of access, third-party audit results, contractor-data handling) would directly address the data-governance concerns that high-stakes AI-lab customers now weigh.

3. Stop redefining the one number you publish

Across April–June 2026 the advertised average ran $99 → $105 → $141/hr — then on 2026-06-30 Mercor relabeled the metric from “average pay” to “average contracted rate” and reset it to $80/hr (a ~43% cut) with no buyer-facing explanation, only to swing it back up to $122/hr two weeks later on 2026-07-14. Changing both the definition and the value of your single public figure, then whipsawing it $141 → $80 → $122/hr inside six weeks, is worse than a steadily rising number: it tells a sophisticated buyer the headline is a marketing dial, not a stable reference, which corrodes whatever anchoring credibility the figure had. The 2026-07-21 and 2026-07-29 captures briefly suggested the swings were narrowing rather than stopping — $122 → $121 → $120/hr, roughly a dollar a week — but that read did not hold: on 2026-08-11 the rate eased to $118/hr, on 2026-08-12 it dropped a further $4/hr to $114/hr (its steepest single-day move since the June reset), and by 2026-08-25 it had kept sliding to $109/hr — its lowest reading since the mid-July rebound, a $9/hr decline across two captures rather than a one-off dip. The same 2026-08-25 capture also showed the underlying “Latest roles” board itself churn unusually hard (seven listings rotated off, seven new ones in), so the volatility isn’t confined to the headline number — the visible role mix behind it is turning over faster than usual too. Five weeks of small drift followed by a sustained multi-week slide is the opposite of a metric settling into a stable, computed blended rate — it is exactly the pattern that should make a buyer discount the figure, and discount it further with each new low. Mercor should say explicitly whether June’s relabel was a real definitional change or a marketing dial, pin down one durable talent-side metric, and pair it with a supply-depth narrative (network size, fill rates) so the number reads as a quality signal rather than a value-metric that can be re-cut at will. Absent that explanation, buyers will keep discounting every figure on the site regardless of how stable any given week looks.

4. Give the data-partnership buyer side the transparency the contributor side already has

The data business is transparent to contributors (“no fees, paid in 2–4 weeks”) but a black box to the AI-lab buyer. As enterprise data-licensing matures, labs will demand provenance, volume, and pricing clarity. Offering a buyer-facing pricing schema for data — even a tiered, volume-based one — would de-risk the largest deals and differentiate Mercor from purely opaque competitors.

The 2026-07-21 rename makes this gap harder to ignore, because the two sides now have separate names — Human data for the lab, Data monetization for the contributor — pointing at one dataset with one undisclosed spread between them. And even the contributor side is less transparent than it looks: the payout formula is “more tools connected = higher payout,” the tool count just went from 34+ to 50+, and there is still no per-tool or per-record figure anywhere. A contributor who connects 20 systems instead of 12 has no way to estimate what that decision is worth before signing, so the widened base raises the stakes of a negotiation they cannot model. Publishing even an indicative payout band per data category would cost Mercor little — contributor payouts are its cost of goods, not its price — and would convert a widening connector list into a genuine reason to connect more.

Key takeaways

1. Decide which side of a two-sided market to make transparent. Mercor makes payouts public and prices private because supply (experts) is its scarce resource and demand (labs) is its margin source. The transparency you choose should solve your hardest acquisition problem, not default to “show the buyer everything.”
2. Opacity is a viable pricing strategy in a young category — temporarily. When no reference prices exist and competitors are equally opaque, a sales-gated, no-rate-card motion lets you price each buyer to willingness-to-pay. But it is a maturity-bound advantage: as the category standardizes, the same opacity becomes a procurement liability.
3. A visible payout is not a price — don’t let buyers conflate them. Mercor’s advertised hourly rate (~$109/hr as of 2026-08-25, its lowest reading since the mid-July rebound, relabeled in June 2026 from “average pay” to “average contracted rate” and swung $141 → $80 → $122 → $121 → $120 → $118 → $114 → $109/hr over roughly eleven weeks) is a wage, not a charge, yet the “contracted rate” wording edges it closer to sounding like a price and lets the two blur in the buyer’s mind. Any marketplace that publishes one side’s economics must be deliberate about whether that number anchors the other side’s expectations — and about how it’s labeled.
4. Margin expansion can come from re-packaging the same supply — including at the label and proof layers. Mercor monetized one expert network as placement, agents, and data — each a higher-margin wrapper. The growth lever was new demand-side SKUs, not new supply, which is far more capital-efficient. The July 2026 rename shows the cheapest version of that move: no new capability shipped, but “Business / Enterprise evals / Data partnerships” became “Enterprise agents / Human data / Data monetization”, turning services into named products a buyer can put in a budget line. The same release widened the data payout base from 34+ to 50+ integrations — a compensation-formula change shipped as a connector list, because Mercor’s payout scales with tools connected. The 2026-08-11 → 2026-08-12 Enterprise-page rebuild is the same logic one layer further down: no new SKU and no new label, just a swap from vanity scale-stats to named, quantified case studies — proof-content curation as a zero-cost credibility lever. Naming a thing, growing the count of what’s countable, and upgrading the proof you show are all pricing-adjacent moves; treat and communicate them as such.
5. Price secrecy raises the stakes of every trust event. When buyers already accept an opaque, asymmetric relationship, anything that breaks trust (a breach, a payment dispute) is disproportionately damaging because the relationship has no transparency cushion to fall back on. Mercor’s breach response shows the cost of under-investing here.

UBP implications

1. The metered unit and the disclosed number can be deliberately different — and the disclosed one can be redefined. Mercor meters buyers on expert-hours / scoped engagements but discloses only the payout side, and across June–August 2026 it both relabeled that figure (“average pay” → “average contracted rate”) and swung it $141 → $80 → $122 → $121 → $120 → $118 → $114 → $109/hr over roughly eleven weeks — including a single-day $4/hr drop on 2026-08-12 (its steepest move since the June reset) followed by a further slide to $109/hr by 2026-08-25, its lowest reading since the mid-July rebound. For UBP strategists, this separates “the value metric you bill on” from “the number you publish” — a powerful but trust-sensitive lever that only works while buyers tolerate opacity; a published number that whipsaws on relabel, briefly appears to settle into a small weekly drift, and then lurches again and keeps drifting to new lows teaches buyers to treat it as noise rather than a signal, and re-earning trust after that costs far more than the volatility itself.
2. Take-rate models resist clean usage-based transparency, and that’s a choice — publishing the unit without the rate is only half a meter. A marketplace could publish its take-rate as a clear percentage meter; Mercor’s refusal to do so shows that opacity is often a pricing decision, not a technical limitation. The same pattern runs on the supply side: the data payout is explicitly a function of tools connected — a real, countable meter that grew from 34+ to 50+ tools on 2026-07-21 — with no per-tool value attached, so a contributor can count their inputs precisely and still not forecast their payout. The UBP lesson is that “we can’t show a rate” usually means “we won’t”, and that a counterparty who sees the unit but not the rate will negotiate against the worst case they can imagine rather than the value you intend.
3. Outcome- and engagement-scoped wrappers can sit on a metered base. Mercor’s enterprise services are engagement-priced on top of an hourly expert-hours base, echoing the broader shift toward outcome-based pricing. Usage-based foundations make it easy to layer higher-margin, value-priced SKUs without rebuilding the meter.

Sources

• Mercor Experts marketplace (talent rates) (accessed 2026-07-29)
• Mercor Data partnerships (accessed 2026-07-29)
• Mercor Enterprise (accessed 2026-07-29)
• Mercor Partner / contact (accessed 2026-07-29)
• Mercor talent help center — Payments (payout cadence, USD denomination, Standard vs 1.0% Instant payout — independently re-verified against the live page, figures unchanged) (accessed 2026-08-15)
• Mercor talent help center — Supported countries for payment (Stripe Connect / Wise rails) (accessed 2026-07-21)
• Mercor blog — Series C announcement (accessed 2026-06-08)

Comparative context across the corpus is available in the pricing blueprint. Funding, take-rate, and breach details cited inline above come from third-party press and analyst coverage (TechCrunch, Wired, Sacra/eesel, Hacker News) recorded in this page’s signal_sources, not from Mercor’s own surfaces.

Bottom line

Mercor is the corpus’s clearest example of a payout-public, price-gated marketplace: it advertises what it pays experts (a ~$109/hr advertised average “contracted rate” as of 2026-08-25 — its lowest reading since the mid-July rebound, $60–$250/hr by role) and data partners — including, as of 2026-07-29, a first-disclosed referral bonus of up to $100K per referral — while keeping the buyer-side take-rate — the actual product price — entirely behind a “Talk to the team” form. The July 2026 rename to Enterprise agents / Human data / Data monetization gave every line a product name a buyer could put in a budget, and the 2026-08-11 → 2026-08-12 Enterprise-page rebuild doubled its named, quantified case studies from three to six — yet none of it, from the renamed lines to the widened 50+ tool payout base to the six proof points, comes with a price. Third parties peg the recruiting fee near 30% and hourly work as cost-plus, but Mercor confirms nothing, so the margin that funds a $10B valuation and a $2B+ run rate stays invisible by design. That opacity is a defensible strategy in a young, undifferentiated category — until a trust event like the March 2026 LiteLLM breach reminds everyone how little transparency the relationship actually rests on.

Want to compare Mercor against other sales-led, usage-based marketplace pricing models? Browse the pricing blueprint.

Pricing timeline : Major events on a vertical axis

Each milestone below corresponds to a public pricing change, product launch, or material adjustment. Major events use a filled marker; minor adjustments use a faded one.

Rate eases to $109/hr; Enterprise page's four offering-step demos refreshed

Aug 2026

The Experts hero 'average contracted rate' fell from $114/hr to $109/hr (roles created up from 349.3K to 401.9K), its lowest reading since the mid-July rebound. Separately, the Enterprise page's Discover/Deploy/Improve/Monetize steps swapped their illustrative product-screen mockups (a workflow-diagnostic dashboard replaced the AI-interview voice demo under Discover; an invoice-run summary replaced the agent-config screen under Deploy; an effectiveness-vs-cost chart replaced the verifier table under Improve; a partially-populated tool-selection/licensing-opportunity panel replaced the blank selector under Monetize) and reworded several 'what you get' bullets under Deploy and Monetize. The same six named case studies and four offering names held over unchanged, and buyer pricing remains fully undisclosed throughout.

Enterprise proof section doubles: three new case studies added, two relabeled

Aug 2026

One day after removing its scale-stats bar, Mercor doubled its Enterprise page's named client case studies from three to six — adding a Workflow Diagnostic (25 candidate AI workflows), a second PE Value Creation Assessment ($20M annual run-rate impact by 2031), and a Production Readiness Evaluation (132 expert-graded HR-agent runs) — while relabeling two existing case-study categories and adding a newsletter signup block. Buyer pricing remained fully undisclosed throughout. Talent-side 'average contracted rate' eased from $118/hr to $114/hr on roles created rising from 339.5K to 349.3K.

Enterprise page rebuilt: scale-stats bar and self-referential case study removed; Agent Benchmarking renamed Agent Optimization

Aug 2026

Mercor rebuilt its Enterprise page: the four offerings are now framed as Discover → Deploy → Improve → Monetize, with 'Agent Benchmarking' renamed 'Agent Optimization' (evaluating agents already in production rather than benchmarking a client's AI products). The prior scale-stats bar ($10B valuation / $2B+ run rate / 100k+ contractors / 400+ employees) and the 'Our first customer was ourselves' AI-support case study are both gone, replaced by three named client engagements (a PE-firm pre-deal assessment citing $7–9M in identified savings, a technology SRE engagement citing a 50%+ reduction in time-to-mitigation, and Mercor's own support operation cited at ~12,000 requests/week), plus new Security & Control and FAQ sections. Buyer pricing remains fully undisclosed throughout. Talent-side numbers moved within routine weekly noise: the advertised 'average contracted rate' eased from $120/hr to $118/hr on roles created rising from 330.6K to 339.5K.

Data-partnership referral bonus disclosed: up to $100K per referral

Jul 2026

The Data page's company-referral module was headlined with its first concrete dollar figure — 'Earn up to $100K per referral' — replacing a prior undisclosed-amount promise ('if it works out, we'll pay you a referral bonus'), alongside a new expandable 'How it works' 3-step flow (refer a company → they onboard → you get paid once their payout settles). Talent-side numbers moved only within routine weekly noise: the advertised 'average contracted rate' ticked from $121/hr to $120/hr on roles created rising from 313.9K to 323.7K; the live 'Latest roles' board rotated its usual weekly set within the existing $60–$250/hr band, and enterprise scale stats and the 50+ integrations figure were unchanged.

Buyer lines renamed to Enterprise agents / Human data / Data monetization; data payout base widens to 50+ tools

Jul 2026

Mercor renamed its buyer-facing lines site-wide — Business → Enterprise agents, Enterprise evals → Human data, Data partnerships → Data monetization — moving from segment/service labels to product nouns, with the APEX research line (Benchmarks, Agents, SWE, Off-the-shelf data) exposed alongside. The Data page's payout base widened from 34+ to 50+ integrations; because Mercor states that connecting more tools raises the payout, this widens the contributor-side compensation formula without disclosing a per-tool rate. The talent headline was flat at $121/hr (from $122/hr) on 313.9K roles created and $4M+ daily payouts, and every renamed line still routes to a 'Talk to the team' form with no rate card.

Headline rate rebounds to $122/hr; revenue run rate doubles to $2B+

Jul 2026

Two weeks after being reset to $80/hr, the Experts hero 'average contracted rate' climbed to $122/hr, while roles created rose to 305.0K and daily payouts ticked up to $4M+. Separately, the Enterprise page's advertised revenue run rate doubled from $1B+ to $2B+ (valuation $10B, 100k+ contractors, 400+ employees unchanged). Individual role listings still span $60–$250/hr and every buyer surface remains a 'Talk to the team' form with no rate card.

Acquires Deeptune to build AI training environments (RL-environment supply)

Jul 2026

Mercor acquired Andreessen Horowitz-backed Deeptune to build training environments and simulators for AI agents (Fortune / SiliconANGLE / The Information, 2026-07-09), its first expansion beyond human-talent matching into the RL-environment supply chain that feeds the same frontier-lab demand its expert hours serve. It is a supply-side/product-surface move, not a pricing move: deal terms were not disclosed and no buyer rate card or take-rate changed.

Headline rate relabeled 'average contracted rate' and reset to $80/hr

Jun 2026

The Experts hero metric was relabeled from 'average pay' to 'average contracted rate' and the figure dropped from $141/hr to $80/hr, while roles created climbed from 258.1K to 287.1K (daily payouts unchanged at $3M+). The relabel suggests a shift from an advertised average pay figure to a blended contracted rate; individual role listings still span $60–$250/hr. Every buyer surface remains a 'Talk to the team' form with no rate card.

Enterprise reorganized into 4 named services; average pay $141/hr

Jun 2026

The Enterprise page was restructured into four named, sales-quoted services (Agent Diagnostics, Deployment, Benchmarking, Data Monetization) and 'Data Partnerships' + 'Enterprise AI' joined the footer nav. The Experts marketplace advertised an average of $141/hr (roles $60–$250/hr), 258.1K roles created, and $3M+ daily payouts — while every buyer surface still routes to a 'Talk to the team' form with no rate card.

Data-partnership business surfaces; average pay $105/hr

May 2026

The standalone Data page (pay-enterprises-to-license-anonymized-operational-data) is archived from May 2026, with 34+ OAuth integrations, SOC 2 Type II, and 2–4 week payouts. Average advertised pay rose to $105/hr; roles created 189.6K (Wayback 2026-05).

LiteLLM supply-chain breach; average pay $99/hr

Apr 2026

A supply-chain compromise of the open-source LiteLLM project (claimed by Lapsus$/TeamPCP) exposed Slack, ticketing, and contractor-interaction data; Wired reported Meta paused work with Mercor. The live Experts page advertised an average pay of $99/hr, 177.5K roles created, and $2M+ daily payouts (Wayback 2026-04; TechCrunch/Wired, 2026-03/04).

Enterprise agents launched as a 'software platform' (WEEK 1–4 build)

Mar 2026

By March 2026 the Enterprise page framed Mercor as an agent-building software platform — a WEEK 1–4 'organizational context graph → agent spec → deploy → iterate' flow, plus an early ACE/APEX benchmark — alongside the $10B / $1B+ run-rate / 100k+ contractor / 400+ employee scale stats. All buyer pricing still sales-gated.

Series C — $350M at $10B valuation (5× in 8 months)

Oct 2025

Felicis-led $350M Series C (with Benchmark, General Catalyst, Robinhood Ventures) quintuples the valuation from $2B to $10B in eight months; Mercor tells investors it is on pace to $500M ARR. Buyer-side pricing remains undisclosed throughout (TechCrunch/CNBC, 2025-10-27).

Series A — $100M at $2B valuation

Feb 2025

Mercor (founded 2023) raises $100M at a $2B valuation as an AI-interview recruiting marketplace placing domain experts onto AI-lab training projects; founders Foody, Hiremath, and Midha are 21. Press reports an average contractor pay of ~$85/hr and ~30k contractors at this stage (TechCrunch, 2025-02-20).