Mercor Review 2026: 5 Million Experts, 30,000 Contracts

Mercor pays the highest published rates in AI training work. Its homepage advertises an average contracted rate of $112 an hour. It also has the narrowest funnel of any platform in this market, and its own numbers show it: Mercor's 2026 security update referred to nearly five million experts on the platform, while its homepage advertises a network of 30,000-plus. Under one percent of the people who sign up are working.

That gap is the honest answer to the question most people arrive with, which is not whether Mercor is legitimate. It is. Founded in 2023 by three Thiel Fellows, it raised $350 million at a $10 billion valuation in October 2025 and supplies expert contractors to frontier labs including OpenAI and Anthropic. It charges experts nothing and takes no cut of your rate.

The question is whether you personally will ever be matched, and two events from the last year change how you should answer it. In November 2025, Forbes reported that Mercor ended a large Meta project and offered contractors replacement work at $16 an hour, down from $21. In March 2026, a supply-chain attack exposed contractor data including identity documents, and Mercor is still notifying affected experts.

This Mercor review covers how the platform works, what it actually pays, who gets matched, both of those incidents in full, and the companies like Mercor worth applying to alongside it.

How does Mercor work?

Mercor screens once, upfront, and then matches you. There is no queue to browse into.

1. Build a profile and upload a resume. Your professional history is the raw material for matching, so the specifics of your practice matter more than a polished summary.
2. Take the AI interview. Roughly 20 minutes. It asks about your background, then pushes into how you reason inside your own field, adapting to your answers rather than reading a fixed script.
3. Your interview is scored and kept. The recording and transcript attach to your profile permanently, and every future project match reads from that score. Retakes are allowed, but the first attempt is the one that matters.
4. Offers come to you. Mercor sends contracts for roles you never applied to, based on your profile against live client demand. Offers are non-exclusive and expire after a set window.
5. Work is hourly and remote. Time is tracked by monitoring software while you work, and rates are published on listings before you accept.
6. You get paid. Mercor's site currently advertises daily payouts, a change from the weekly cycle most contributors describe, paid in USD and converted to local currency.

Two things Mercor genuinely does better than its competitors: it publishes a rate on nearly every listing, which most of this market does not, and it is free for experts with no cut taken from your pay. Most listings never name the client. Your contract is with Mercor and the lab stays anonymous.

The number that explains why you have not been matched

Nearly five million experts on the platform. A network of 30,000-plus. Those are both Mercor's own figures, from its security update and its homepage.

Registered and contracted are different numbers, and the distance between them is the single most useful fact about this platform. It means:

• Passing the AI interview is not acceptance. It produces a score. The score sits idle until a client needs your exact profession.
• Matching is demand-driven, not effort-driven. No amount of profile polishing creates client demand for your field. This is the same structural ceiling every platform in this market has, and Mercor is the most explicit version of it.
• Scarcity is the product. Mercor sells labs access to judgment they cannot hire quickly. A practising radiologist, a securities attorney or a staff engineer clears the bar faster than a strong generalist, because the generalist is not scarce.

If you have a licensed profession or deep specialist practice, Mercor is probably the highest-paying platform available to you. If you do not, it is the one most likely to leave you with a completed interview and silence.

What Mercor pays in 2026

• Advertised average: $112 an hour, on Mercor's own homepage. The highest published figure in this market.
• Expert contracts: medicine, law, finance and senior engineering regularly post above $100 an hour, with some roles priced per engagement rather than hourly.
• Generalist evaluation work: far lower, and this is where the average breaks down.

That average is not a typical rate. It is pulled up by scarce professions, and it tells you nothing about what a generalist role will pay. The November 2025 Forbes reporting made the spread concrete: contractors who had earned up to $60 an hour on other Mercor projects were offered replacement work at $16.

Mechanics worth knowing before you accept anything:

• The rate belongs to the project. It is published upfront and it is not negotiated.
• Hours are tracked by software. Your paid time is what the tracker records.
• Offers expire. Non-exclusive, time-limited, and declining one does not remove you from matching.

The Meta project pay cut, in full

In November 2025, Forbes reported that Mercor told thousands of contractors they would no longer work on a large Meta project codenamed Musen, citing project scope changes. More than 5,000 people had worked on it at its peak.

Contractors were then offered a similar replacement project, Nova, at $16 an hour. The Musen rate had been $21, making the new rate roughly 24% lower. Workers were reportedly told the new project offered greater weekly engagement through higher hour caps and steadier volume. Contractors who spoke to Business Insider called the rate drastically low, particularly those who had earned up to $60 an hour on other Mercor work. One told Forbes they deserved notice or warning.

Mercor's position is that this is the normal course of business with temporary workers and the shifting needs of client projects and budgets. That is a defensible description of how contract work functions. It is also the entire risk of the model, stated plainly: your rate is set by a client contract you cannot see, and it can be replaced by a cheaper one without notice.

Contractors interviewed by several outlets across 2025 and 2026 described a stressful working environment, thin management and declining pay. Read that alongside the Trustpilot picture below rather than instead of it.

The March 2026 breach: what contractors should know

This is the part of any Mercor review that matters most, because Mercor holds identity documents.

What happened. In late March 2026, attackers compromised the open-source LiteLLM package and published malicious versions to PyPI. The compromise reportedly began upstream, with a widely used vulnerability scanner altered to point at a malicious release, and publishing credentials extracted from LiteLLM's build pipeline. Mercor installed the affected versions.

What was claimed. The attackers claimed roughly four terabytes of data. Reporting described the haul as including platform source code, a user database, and storage containing video interviews, passport scans, Social Security numbers and identity verification documents. Mercor confirmed the incident publicly on 31 March 2026.

What Mercor says. It engaged Google's Mandiant, outside security firms and law enforcement. Its published finding is that of its nearly five million experts, only a very limited subset had sensitive information affected, and that it found no evidence of fraudulent use. It said no employee data was compromised and that customer impact was minimal. It began notifying affected experts directly from 25 June 2026, offering TransUnion identity protection. Remediation included auditing third-party dependencies, rotating credentials across cloud platforms and GitHub, penetration testing and 24/7 managed detection.

Where that leaves you. Multiple class actions followed, including one brought by a contractor who had submitted a W-9 with personal information. The gap between the attackers' claims and Mercor's characterisation is not resolved, and you should not expect it to be resolved quickly.

Practical steps if you sign up or already have:

• Assume the identity documents you upload are a real asset worth protecting, not a formality.
• Consider a credit freeze. It is free in the US and costs you almost nothing in inconvenience.
• Use a unique password and enable every authentication factor offered.
• If you worked through Mercor in the affected period, check whether you were notified and take the identity protection on offer.

None of this makes Mercor uniquely unsafe. Its response was faster and more detailed than most companies manage. It does mean the decision to upload a passport here should be a decision rather than a reflex.

Is Mercor legit? What the review data actually shows

Mercor holds 4.4 out of 5 on Trustpilot across roughly 665 reviews. The average is less informative than the distribution: around 73% are five-star and around 12% are one-star, with very little in the middle.

That bimodal shape is the most honest summary of this platform anyone can give you. Reviewers who get matched describe smooth onboarding, paid training time, communicative project teams and payments that arrive on schedule. Reviewers who do not describe applications that go nowhere, projects cancelled before work begins, automated support with no human behind it, and no explanation for rejection. Both groups are describing the same company accurately.

The complaint that appears on both sides is instability: projects ending suddenly and terms changing mid-contract. That is consistent with the Forbes reporting, and it is the thing to plan around.

Who gets matched on Mercor

The verticals with the most live demand: physicians and medical specialists, attorneys, finance professionals, senior software engineers, and PhDs across the sciences. Generalist writing and evaluation roles exist at lower rates.

The pattern in who actually gets contracts: years of real practice beat credentials alone, and credentials beat enthusiasm.

How to pass the AI interview. Candidates fail by narrating their resume. The interview is scoring how you think, so walk through a real decision from your own practice the way you would teach it: what you considered, what you ruled out, why you chose what you chose. Be specific, be complete, and speak from work you actually did. Retakes are allowed, but the score attaches to your profile and every match reads from it, so prepare for the first attempt as though it is the only one.

Companies like Mercor

The full field is ranked in the best AI training jobs roundup. The short version:

• micro1. The closest comparison. AI-vetted network, skill-based screening rather than credential-based, roles from $40 to $60 an hour for engineering and $100 to $175 for senior, medical, legal and PhD-level work. The better door if your expertise is real but your credentials are not formal, and its revenue is growing where Mercor's project volume is not. See the micro1 review.
• Alignerr. Labelbox's expert network, weighted to coding and language work, advertising an $80 average. Applications convert far more often than Mercor's matching does, but read the payment model first: the Alignerr review explains why that matters.
• Outlier. Scale AI's marketplace, with rates published by domain: roughly $15 to $30 for general rating, $30 to $50 for coding and math, up to about $100 for legal and medical review. Its 2026 problem is volume rather than rates, covered in the Outlier AI review.
• Terac. Paid research studies for verified professionals, priced per approved session at $100 to $350. For a sub-hour interview that outruns nearly every hourly rate here.
• DataAnnotation. The lowest entry bar, roughly $20 to $40 an hour, gated by one unpaid assessment with a single attempt. See the DataAnnotation review.
• Remotasks. Scale AI's volume annotation platform, still running, mostly $5 to $15 an hour and priced by country. The Remotasks review covers its verified status.

Should you sign up for Mercor?

Yes, if you are: a licensed professional in medicine, law or finance; a senior or staff engineer; a PhD in an active research field; or anyone whose day job involves judgment that takes years to acquire. Your rate here will beat every other platform in this market, and the 20-minute interview is a cheap option to hold.

Probably not, if you are: looking for guaranteed volume, dependent on a predictable weekly figure, or a strong generalist without a scarce credential. You can pass the interview and never hear anything, and that is the modal outcome rather than a failure on your part.

Either way: do not treat an accepted Mercor profile as an income. The Musen contractors had one, and it ended with a scope change and a 24% lower replacement rate.

The pattern across every platform in this market

Mercor pays the most and matches the fewest. Outlier publishes honest rates and lost its volume when Meta bought into Scale AI. DataAnnotation has the easiest door and the lowest ceiling. Remotasks is still running and barely maintained.

Every one of them is a queue owned by someone else's client contracts, and every one of them goes quiet. Contributors who earn steadily are qualified on two or three at once, so a cancelled project on one is covered by another.

Doing that used to mean checking six boards with six sets of titles, pay conventions and signup flows. Skillora's AI training jobs board collects live listings from Mercor, micro1, Alignerr, Outlier, DataAnnotation and Terac into one feed, refreshed daily, filterable by pay, skill and experience level. Every listing links back to the marketplace that posted it, you apply and get paid by them directly, and browsing is free. There is a Mercor jobs page with current openings and published rates.

One more thing specific to Mercor. Its entire screening is a 20-minute AI interview, and your score follows you permanently. If you have not sat an interview in years, and especially if your experience is task queues rather than conversations, running that format once with scored feedback beforehand costs an evening and changes what your profile is worth for as long as you hold it.

Verdict

Mercor is legitimate, pays the best published rates in AI training work, and is the hardest platform here to actually get onto. Under one percent of registrations turn into contracts, and that is a demand problem, not a quality problem.

Take the interview if you have a scarce profession, because the option is cheap and the ceiling is high. Prepare properly, because the score is permanent. Take the identity-document question seriously after March 2026. And qualify somewhere else too, because a Mercor contract is a client's budget decision away from ending.