Sepal AI Review: Is It Legit, and Who Owns It Now?

Is Sepal AI legit? The short answer

Yes, Sepal AI was a legitimate company, and it no longer operates on its own. Mercor, the AI training-data company, acquired it in February 2026. People who search "Sepal AI" today tend to be one of three groups: experts who did paid work on Sepal projects, AI teams that bought evaluation data from it, and owners who heard that Sepal pays for data.

Each group needs a different answer. Experts and customers now deal with Mercor. Owners should know that Sepal paid people for expert work, not companies for their records, and that Mercor runs a separate, published program for company data. The sections below set out what Sepal made, what changed with the acquisition and where each group goes now.

What Sepal AI did, and for whom

Sepal's Y Combinator profile says "Sepal AI was founded in 2024 by engineers and operators from Vercel and Turing", that it "raised several million dollars from leading investors" and that it "already count[s] multiple Fortune 500s and top AI research labs as paying customers", a claim the company made about itself (Y Combinator). The profile lists a team size of 15 and now reads "Summer 2024 Acquired".

Both co-founders came from Turing's model-training business. Robi Lin "Scaled Turing's LLM trainer business line from 50 to 800+ onboarded developers in 5 months", and Kat Hu "Built Turing's Foundational LLM trainer business GTM" (Y Combinator). Lin later wrote that they started Sepal because "AI labs were running out of data, and improving models further required producing entirely new kinds of datasets" (Robi Lin).

Sepal described its product as a "Cloud-Native Agent Dataset Factory" for evaluation and training data, and said: "We partner with the world's leading AI labs and enterprises to help their models get better at the tasks people actually want them to do" (Y Combinator). Orrick, the law firm that advised Mercor, called it "a data development platform specializing in creating high-quality training data, evaluation benchmarks, and Reinforcement Learning (RL) environments for frontier Large Language Models (LLMs)" (Orrick).

Its launch post gives concrete examples. A "Cell and Molecular Biology Benchmark: An original benchmark to evaluate complex reasoning across models." A "Finance Q&A + SQL Eval: A Golden Dataset to test the ability of an AI agent to query a database and produce human-expert-level answers to complex finance questions." And "Uplift Trials & Human Baselining", which it described as "End to end support for conducting secure in-person evaluations on model performance." Y Combinator also lists a news item: "Sepal AI supports the launch of Anthropic's Claude 3.7 Sonnet with Pre-launch Testing" (Y Combinator).

The people doing the work were experts. At launch Sepal claimed "a network of 20k+ experts across STEM and professional services" (Y Combinator), and at the sale Lin wrote of "working with thousands of experts across scientific and professional domains" (Robi Lin). Sepal paid those experts. Its customers paid Sepal. No company was paid for its records.

Source: Y Combinator: Sepal AI profile

Who owns Sepal AI now: the Mercor acquisition

Mercor announced the deal in early February 2026: "Today, we're welcoming Sepal AI to Mercor." It called Sepal "an exceptional team working at the intersection of human data, reinforcement learning environments, and specialized research projects" and said "They have already proven to be a trusted partner to frontier AI labs and early enterprise teams" (Mercor). Sepal's co-founder posted the same day: "Today we're excited to share: Sepal AI has been acquired by Mercor" (Robi Lin). Neither post gave a price.

When Origin checked on 10 November 2026, www.sepalai.com, the address still listed on Sepal's Y Combinator profile, returned a 404 (page not found) error. That fits a brand folded into its buyer. It also means a live site or email address using the Sepal name is not something we could confirm as Sepal's, so check any such contact with Mercor before you reply.

What the acquisition changed for customers, experts and companies

For AI labs and enterprise teams that bought from Sepal, the contract counterparty is the question. Ask Mercor which entity now holds your agreement, whether the confidentiality terms carried over unchanged, and who owns the datasets and benchmarks Sepal delivered to you. If you commissioned a private benchmark, confirm in writing that it stays private to you now that Sepal sits inside a larger supplier that serves other labs.

For experts who worked on Sepal projects, the work and the relationships now run through Mercor. Your confidentiality duties under a Sepal agreement do not lapse because the company was sold, so keep treating project material as confidential. Keep copies of your Sepal agreements and payment records, and raise any unpaid invoice with Mercor's support rather than an old Sepal address.

For companies that hold records, nothing changed, because Sepal never bought company records in the first place. The route that exists today is Mercor's own data program, set out below, or one of the other buyers that publish terms.

Why Mercor bought Sepal AI, and what it says about the market

Mercor bought Sepal for its people and its environment work. Its announcement said Sepal's talent and technology would strengthen its work "to advance long-horizon research and train AI models against the most complex tasks that humans perform" (Mercor). Sepal was not the only purchase. Fortune reported that Deeptune, which builds simulated work environments, "raised a $43 million Series A led by Andreessen Horowitz in March and agreed to be acquired by Mercor four months later" (Fortune).

The same Fortune piece noted that "Scale AI wrote in February that nearly half of its new data-training projects involve reinforcement-learning environments". Large data suppliers are buying the small teams that build these environments rather than building every one themselves.

For a company that holds records, this consolidation has two effects. Fewer, larger buyers means fewer independent quotes, so it pays to compare the programs that remain. And environments need realistic material: real workflows, real documents, real decision trails. That demand is why Mercor, micro1, Handshake and others now pay companies for records.

Source: Fortune: nine-person Halluminate raises $30 million (Oct 2026)

Sepal AI for experts: where the work went

Sepal's expert work was specialist work: writing hard questions in a field you know, grading model answers against an expert standard, and taking part in supervised trials. If you did that work and want more of it, the closest match is the expert side of Mercor, which absorbed Sepal's team. Our Is Mercor legit? review covers how Mercor pays experts and what to check in a contract, and Mercor competitors lists the other firms that hire for similar projects.

If an invitation still arrives under the Sepal name, treat it with care. Sepal's old address did not load when we checked, so a current offer should come from Mercor. Do not send identity documents or bank details in reply to an unsolicited message; start a new conversation through Mercor's own website instead.

Does Sepal AI buy company data? What to do instead

We found no Sepal program that bought records from companies. Its model was paid expert work sold to AI labs and enterprises, not data licensing. A company that wants to license its records should look at the programs that publish terms.

Mercor, which now owns Sepal, runs one. It says "We license it, package it, and pay you for it", that "Compensation is based on data volume, the tools you connect, and the depth of your data", and that "Payment is issued via wire transfer upon delivery of the anonymized dataset to the buyer", which "typically occurs within 2-4 weeks of the signed agreement" (Mercor). Step-by-step terms are in our Mercor data monetization program guide.

Sepal's own projects hint at what this market values. Its finance test set asked an agent to query a database and answer finance questions at expert level. In our reading, a company with years of real reporting queries, the questions that prompted them and the answers that went to management holds the raw material for that kind of task, which an expert network has to invent from scratch.

Compare before choosing. Company data buyers compared sets Mercor against micro1, Handshake and Turing, our RL environment companies tracker shows which environment builders pay for records, and our directory of AI data companies covers the wider market. Before you sign with any of them, run the checks in how to vet an AI data buyer.

Source: Mercor: monetize your data

Next step

If you are weighing an offer for company records, compare it with what similar records have fetched elsewhere before you sign.

Not sure whether a buyer or an offer is sound? Origin checks who the buyer is, what the license allows and how the price compares with published deals before anything is signed. Get a free estimate and a second opinion.

Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.