Mercor vs Turing: How They Compare for AI Training Work (2026)

Short answer#

Mercor and Turing both help AI labs and enterprises get the human expertise and data that frontier models need. The overlap is large: each runs a network of experts, each builds evaluations and reinforcement learning (RL) environments, and each has a public offer for businesses that want to monetize internal data.

The difference is mostly where they came from and how they are structured. Mercor was founded in 2023 as an AI hiring platform and became a marketplace for domain experts, with a 2025 valuation of $10 billion. Turing was founded in 2018 as a remote developer platform, expanded into AI training in 2022, and now runs three businesses side by side. Mercor has had a difficult 2026 on security and legal fronts. Turing has been quieter in the news and has not announced funding since 2025.

This page is written as of October 2026. Funding, valuations and revenue figures below are mostly company-reported or come from press reports citing the companies, and we flag where sources conflict.

At a glance#

Company background#

Mercor#

Mercor was founded in 2023⁴ by Brendan Foody, Adarsh Hiremath and Surya Midha, who received Thiel Fellowships in March 2024, according to The Harvard Crimson. The company initially focused on recruiting software engineers and technology professionals for startups.⁷ TechCrunch described the same arc: an AI-driven hiring platform that pivoted to supplying scientists, doctors and lawyers for model training.⁸

Mercor's funding history, as reported:

• February 2025: a $100 million round led by Felicis at a $2 billion valuation, up from $250 million the previous September, according to Bloomberg. The company said it was profitable at the time.⁶
• October 2025: a $350 million Series C at $10 billion, led by Felicis, with Benchmark, General Catalyst and Robinhood Ventures participating, per TechCrunch.⁸
• July 2026: Bloomberg reported early talks at about $20 billion, and Mercor told investors it had received a term sheet at that valuation, per TechCrunch. As of the sources we read, we found no confirmation that a round has closed, so treat $20 billion as unconfirmed.¹⁰

On scale, TechCrunch reported in October 2025 that Mercor said it paid contractors more than $1.5 million per day, and that it apparently told investors it was on track for $500 million in annual recurring revenue (ARR).⁸ In July 2026 CEO Brendan Foody said on X that the annualized revenue run rate had passed $2 billion, according to TechCrunch.¹⁰ That figure comes from the CEO's own post and is not audited.

Turing#

Turing was founded in 2018 by Jonathan Siddharth and Vijay Krishnan.³⁵ It began as a platform that used AI to source, vet, hire and manage remote engineers. In December 2021 it raised $87 million at a $1.1 billion valuation led by WestBridge Capital, and TechCrunch listed customers including Johnson & Johnson, Coinbase, Rivian, Dell and Disney at that time.¹⁵

TechCrunch reported that in 2022, OpenAI approached Siddharth, because its researchers had found that code in training data improved reasoning, and asked Turing to generate that code. That began Turing's AI training business. Siddharth pushed back on calling it a pivot, and TechCrunch reported that the original sourcing business still generated substantial revenue.⁹

In March 2025 Turing announced a $111 million Series E at a $2.2 billion valuation, led by Malaysia's sovereign wealth fund Khazanah Nasional Berhad, bringing total funding to $225 million. TechCrunch said the round was priced at $167 million ARR, that ARR was about $300 million at the time of writing, and that Siddharth said the company had been profitable for about a year.⁹ Turing's own announcement confirms the round size, valuation and total funding.⁵

One note on dates: Turing's own blog post carries a June 2025 date, while TechCrunch dated the news March 2025. We use TechCrunch's March 2025 date, and we found no Turing funding announcement since.

What each company does#

Mercor's business#

Mercor describes itself as "organizing human intelligence to power the AI economy" and compares its expert network to the driver network of a ride-hailing service.¹¹ Its offerings, per its own site and blog:

• Expert work and human data. Experts in fields such as law, accounting, software engineering, banking, consulting and medicine take on projects for AI labs and enterprises.¹
• Evaluation and benchmarks. Mercor publishes the APEX benchmark family, which tests AI models on professional work, with named partner companies on the homepage including Harvey, Ramp, Cognition and Sierra.¹ In October 2026 it said it is building a research team to study measuring AI on real work, producing better training data, and post-training models with that signal.¹⁶
• Enterprise agent services. Its enterprise page describes four stages: diagnosing AI opportunities, deploying agents around a client's workflows, optimizing production AI, and monetizing data.¹⁷ The page lists client engagements and claimed results, which are marketing claims we did not verify independently.
• RL environments. In July 2026 Mercor announced it would acquire Deeptune, which it described as a company building environments for reinforcement learning whose team has rebuilt hundreds of business applications, from spreadsheets to Salesforce. Deeptune had raised a $43 million Series A led by Andreessen Horowitz.¹⁸
• Data licensing. Covered in the data seller section below.¹³

Turing's business#

Turing's company page says it provides data, evaluations and RL environments to frontier AI labs and builds agentic AI systems for Fortune 500 enterprises. The page also states 5M+ domain experts, 1M+ curated tasks, and that Turing works with "9 of 9 frontier labs" (a company claim).³

Turing's own announcements describe three focus areas:⁵

1. AGI Advancement. Acting as a research accelerator for frontier labs, with data and post-training programs.
2. Turing Intelligence. Building enterprise AI systems by fine-tuning models on a client's proprietary datasets.
3. Deliver Talent. Placing vetted engineering talent with AI companies and enterprises, the original business.

On the frontier side, Turing's NeurIPS 2025 page lists a data catalog covering coding, STEM, multimodal, domain-specific (finance, medical, legal and economics) and robotics work, plus benchmarks such as SWE-bench++, built from real GitHub issues and validated fixes, and RL environments for transactional, support and project-management tasks.¹⁹ Turing's site also advertises off-the-shelf datasets that visitors can request samples of.² A Turing Frontier page describes an expert network spanning engineering, finance, law, medicine and STEM, claims 250+ domains and a 5M+ expert pool, and routes clients through a "Hire experts" form.²⁰ That page did not describe vetting steps or pay in the version we read.

On the enterprise side, Turing's LLM Lab page describes fine-tuning models on a client's proprietary datasets inside a sandbox, with support from engineers experienced in supervised fine-tuning and reinforcement learning from human feedback.²¹ That is a different relationship from the one described below under data sales. In the enterprise line, the client's data is used to build that client's own system. This is our reading of the pages, and neither page addresses the other use.

How they source people and data#

Mercor#

Plaintiffs in one lawsuit allege that applicants must complete AI interviews and automated screening,²² and Mercor describes its experts as vetted. The company said in March 2026 that its network exceeded 4 million vetted experts with more than 1.9 million referrals.¹¹ In June 2026 it referred to "nearly five million experts."¹² The homepage separately states 30k+ experts, and in October 2025 TechCrunch cited more than 30,000 experts on the roster earning over $85 per hour on average.¹⁸ These figures are not directly comparable. The larger numbers appear to count people who have signed up or been vetted, and the smaller ones appear to count experts actually working, but Mercor has not defined the terms in the pages we read.

The homepage lists an average contracted expert rate of $107 per hour and daily payouts above $4 million, with example listings such as an oncologist at $350 to $450 per hour and a sales and marketing expert at $60 to $70 per hour.¹ Mercor's earnings model for itself, per TechCrunch, is an hourly finder's fee plus a matching rate on expert work.⁸

Pay is not uniform. Forbes and Business Insider reported in November 2025, as summarized by People, that contractors on a content-review project, paid $21 per hour, learned the project was canceled and were offered a replacement project at $16 per hour. Mercor said the claims were inaccurate and described the work as temporary and project-based.²³

Turing#

Turing's expert network grew out of its developer vetting platform. In 2021 it reported about 1 million developers from 140 countries in its candidate pool,¹⁵ and in 2025 TechCrunch cited about 4 million coders worldwide.⁹ Its current company page claims 5M+ domain experts.³ The Turing Frontier page says clients receive "a matched set of vetted experts," but the version we read did not describe the vetting or pay.²⁰

For datasets, Turing says expert-verified data comes from its expert network through auditable pipelines with human-in-the-loop quality control.¹⁹ The separate source of data, businesses selling existing code and records, is covered in the data seller section.

Customers and partners#

Only publicly confirmed names are listed here.

• Mercor. Co-founder Adarsh Hiremath named OpenAI as a client in a 2025 Harvard Crimson interview.⁷ Mercor's homepage names Harvey, Ramp, Cognition and Sierra as benchmark partners, which is not the same as being data customers.¹ In its June 2026 incident update, Mercor said all frontier labs had increased their work with it in recent months, without naming them.¹²
• Turing. Turing's homepage shows logos for Anthropic, Gemini, Nvidia, Snowflake and Augment, and the NeurIPS page adds Character.ai.²¹⁹ Logos are not contract descriptions. The 2021 Series D article named Johnson & Johnson, Coinbase, Rivian, Dell and Disney as talent-platform customers at that time.¹⁵

Controversies and legal issues#

All items below are allegations or reports unless stated otherwise. We cite the source in each case and note when the source is a plaintiff law firm.

Mercor#

Scale AI lawsuit (September 2025). Scale AI sued Mercor and former Scale employee Eugene Ling. Scale alleged Ling took more than 100 confidential documents about customer strategies and pitched Mercor to one of Scale's largest customers before leaving. Mercor co-founder Surya Midha denied that Mercor used any Scale data and said Mercor had offered to have Ling destroy the files. Ling said he had no nefarious intent. TechCrunch's report did not give the outcome, and we did not find a resolution.²⁴

Contractor pay dispute (November 2025). Described above.²³

March 2026 security incident. Mercor disclosed on March 31, 2026 that it was affected by a supply-chain attack involving the open-source tool LiteLLM.¹²²⁵ TechCrunch reported that the tool harbored credential-harvesting malware for 40 minutes, and HR Dive, citing TechCrunch, said Mercor was one of thousands of companies affected.²⁵²⁶ A hacker group claimed to have taken 4 terabytes of data, including source code and API keys, but Mercor had not commented on whether the data was authentic.²⁵ Per TechCrunch, citing Wired, Meta paused its Mercor contracts indefinitely, and OpenAI said it was investigating its exposure without ending its contracts.²⁵ In a June 25, 2026 update, Mercor said its investigation with Google's Mandiant and others was complete, that "only a very limited subset" of nearly five million experts had sensitive information affected, that customer impact was very limited, and that it was notifying affected experts and offering identity protection through TransUnion.¹² The two accounts differ in tone, and we could not independently verify either claim about scope.

Class actions. HR Dive reported that several plaintiffs filed class actions beginning April 1, 2026 in the Northern District of California over the breach, alleging negligence, unjust enrichment, breach of implied contract, breach of privacy and violation of California's Unfair Competition Law.²⁶ A later putative class action, announced April 21, 2026 by plaintiffs' counsel Hall Attorneys, alleges that Mercor used AI interviews and automated screening and treated workers as independent contractors rather than employees. The release notes the complaint contains allegations only.²² Separately, White v. Mercor.io Corp., No. 6:26-cv-00201, is a proposed class action over worker misclassification filed May 8, 2026 in the Northern District of Texas. Law360 reported in June 2026 that Mercor asked the judge to send it to arbitration.²⁷ A summary by a Littler Mendelson attorney says the plaintiff seeks to represent about 30,000 experts and asserts ERISA and Form 1099 claims.²⁸ TechCrunch reported that Mercor declined to comment on the contractor lawsuits.²⁵

Turing#

We searched for lawsuits, regulatory actions and major controversies involving Turing and did not find any reported in news sources. The Littler Mendelson summary of misclassification suits against AI firms that we read did not mention Turing.²⁸ Absence of reports is not proof of absence, and AI data vendors across the industry are facing worker-classification claims.

If you are selling business data#

This is the section most relevant to this site's readers. It is based only on what each company publishes and what independent reporting shows.

Mercor's data licensing program#

Mercor's data page says it licenses business workflow data to frontier AI labs, packages it, and pays the business that supplies it. The process, as described, has four steps: choose which tools to share, connect accounts from 100+ supported tools, have the platform extract and de-identify the data, and receive payment once processing is complete.¹³ Key stated terms:

• Participation is described as opt-in on Mercor's enterprise page, and the data page says "No data leaves your systems until you approve it."¹⁷¹³
• The page says no AI lab buyer or third-party vendor sees raw data, which is Mercor's own claim.¹³
• Mercor says it holds SOC 2 Type II certification and manages encryption keys in AWS KMS.¹³
• Pay depends on data volume, the tools connected and data depth. The page does not publish a rate card.¹³
• Mercor lists no upfront costs, setup fees or deductions, and says payment typically comes within 2 to 4 weeks of the signed agreement.¹³
• Target industries include legal, healthcare, finance and insurance, professional services, manufacturing, retail and e-commerce, real estate and logistics.¹³

Mercor's enterprise page lists the same offer under "Monetize," noting that it is always opt-in.¹⁷

Mercor also bid on a large data archive. When Spirit Airlines went through bankruptcy in 2026, Google won the auction for its corporate data with a $10 million bid, and Mercor bid $7.5 million and lost, according to Time. The dataset included 100 million emails, 500 million Microsoft Teams messages and around 30 million lines of code. A flight attendants' union objected on privacy grounds, and Time reported that a judge was set to rule on the objection on September 9. Time also reported that micro1 submitted a higher $12.5 million bid after the auction closed.²⁹ MIT Technology Review also reported that Google outbid Mercor in August 2026.³⁰ We could not find a report of the final ruling, so the outcome is unclear as of this writing.

Turing's Project Lazarus#

Turing's Project Lazarus page describes Turing acquiring operating records from companies, including team communications, source code and reviews, tickets, wikis, runbooks, schemas and decision history, across software, finance, customer support, sales, legal and operations categories.¹⁴ It targets "operating and winding-down companies." The stated process:¹⁴

1. Book a call, scheduled within one business day, under NDA.
2. Turing reviews a representative sample, usually within a week.
3. The seller receives one valuation and one contract, with no retainer or recurring commitment.

The page advertises a single payment of up to $1 million that scales with how complete, rare and well-preserved the record is, a referral bonus of up to $10,000, de-identification of personal data and secrets before data reaches a model, air-gapped processing, SOC 2 Type II, and "clean rights transfer."¹⁴

In the version we read, the page listed eight FAQ questions, including who owns the data, how the price is decided, and who is eligible, but the answers did not appear in the content we retrieved.¹⁴ A prospective seller should therefore not assume the terms and should ask for them in writing.

What the comparison means#

Other points sellers should weigh:

• Type of data. Turing's page emphasizes code, tickets and engineering records, though it lists other categories. Mercor's page emphasizes workflow data from business software such as the 100+ tools it supports.¹⁴¹³
• Privacy risk. Both rely on automated de-identification. The Spirit Airlines dispute shows that unions and regulators may challenge whether anonymized employee data can be reconstructed.²⁹
• Vendor risk. A buyer's own security record matters when you hand over access. Mercor's incident involved contractor data, and a hacker group claimed to have taken source code and API keys, which Mercor had not confirmed; Meta reportedly paused work.²⁵ Turing had no comparable public incident that we found.
• Price transparency. Mercor's page says compensation depends on volume, tools and depth.¹³ Turing's says up to $1 million.¹⁴ Neither is a quote. Reported prices for codebases from other sources vary widely, for example about $25,000 for one startup's code, per MIT Technology Review.³⁰

Strengths and weaknesses#

Mercor#

Strengths:

• Very high reported growth, with a $10 billion priced valuation and a CEO-stated run rate above $2 billion.⁸¹⁰
• A broad, specialized professional network and its own benchmarks, with an in-house research team announced in October 2026.¹¹⁶
• An explicit, opt-in data licensing program with stated payment timing.¹³

Weaknesses and risks:

• The 2026 security incident, the reported Meta pause, and multiple lawsuits.²⁵²⁶²²
• Contractor relations: pay changes, project cancellations and misclassification claims.²³²⁷
• Rapid change: valuation and revenue figures move fast, and the $20 billion round is unconfirmed.¹⁰

Turing#

Strengths:

• Longer operating history and three lines of business, which spreads reliance across customer types.⁵
• TechCrunch reported in March 2025 that the CEO said the company had been profitable for about a year, with ARR of around $300 million.⁹
• An enterprise offering for companies that want to keep and use their own data, plus a purchase program for those that want to sell it.²¹¹⁴

Weaknesses and risks:

• Little public detail on expert pay, vetting steps, or sale terms. Several Turing pages we read had blank FAQ answers.²⁰¹⁴
• Last funding information is from March 2025, so figures are older than Mercor's.⁹
• Customer names are shown as logos, with no contract details.²

Do not let one buyer set your price. Any single buyer's published range, program page or first offer is that buyer's number, not the market's. The sensible move is to get competing offers on the same data, compare scope, exclusivity, indemnity and payment timing side by side, and have someone independent look at the terms. Our guide to selling direct vs using a broker explains how, and if a buyer is already talking to you, we review offers free.

Who each suits#

Mercor suits:

• AI labs and enterprises that want credentialed professionals across many occupations for evaluation and training data.
• Enterprises looking for a vendor that bundles agent diagnostics, deployment and evaluation.
• Companies with data in common business tools that want to explore an opt-in licensing program, after due diligence on the incident.

Turing suits:

• Labs that need coding-heavy datasets, SWE-style benchmarks and RL environments.
• Enterprises that want a custom model fine-tuned on their own data.
• Companies with code repositories and engineering records, especially if winding down, that want a single-offer process.
• Teams that need engineers placed directly, since talent placement remains a business line.

Summary#

Mercor and Turing overlap on the supply side of AI training: both provide experts, evaluations and RL environments to frontier labs, and both operate at the multi-million-expert scale they claim. Mercor is newer and faster growing by reported numbers, and its 2026 has been dominated by a security incident, reported client caution and litigation. Turing is older, reports profitability, and has three business lines, but publishes less about pay and process.

For businesses holding internal data, the two companies' public offers are the notable point: Mercor frames its program as opt-in licensing of anonymized workflow data, and Turing frames Project Lazarus as a purchase of code and operating records. Both leave key commercial terms to private negotiation.

Key points#

• Valuation and funding: Mercor $10 billion (October 2025, talks reported at about $20 billion in July 2026), Turing $2.2 billion (March 2025).
• Origins: Mercor from AI-driven hiring (2023), Turing from remote developer staffing (2018).
• Both offer ways for businesses to monetize data, with undisclosed pricing.
• Mercor's 2026 incident and lawsuits are the main legal and reputational difference.
• Ask for ownership, exclusivity and de-identification terms in writing before any data leaves your systems.

Verdict#

If you are a buyer of expert work, Mercor offers the broader professional range and more public benchmark activity, while Turing offers a coding and research focus with an enterprise AI option. If you are a possible seller of business data, the two programs differ in type of data and process, and neither publishes enough to compare offers without a direct conversation and a written contract. Do your own diligence on security, rights and privacy, and do not rely on either company's marketing alone.