- Abaka AI is an operating AI data company: it sells off-the-shelf datasets, expert annotation, model evaluation and RL environments, and lists its open roles in Palo Alto, California.
- It announced $8 million in funding in May 2026. A Forbes profile of co-founder Jack Lin from 2024 put total funding at $20 million, and neither source gives a current total.
- Project Reborn pays companies for private codebases and for company operating data such as docs, chats, tickets and SOPs. The headline figures are ceilings: up to $1M per repo and up to $5M per company.
- Abaka publishes its process (a human reply within 48 hours, a written offer within 14 days, two-part payment, no upfront fees) but not a sample agreement or the AI labs that receive the data.
Origin Data Partners is an independent data partner that helps business owners value and license their company records: the company pays Origin nothing, and a buyer may pay Origin when a deal completes.
Is Abaka AI legit?
Yes. Abaka AI is an operating company with a public product line, published dataset prices and open roles, and in May 2026 it announced $8 million in new funding (Yahoo Finance). People who search for it tend to be one of three groups: AI teams shopping for datasets or annotation, experts looking for paid work, and company owners who saw its Project Reborn offer of up to $1M for an old codebase.
Each group needs a different answer, so this review takes them in turn: what Abaka sells, what is on record about its funding and ownership, how it works with experts, and what its company data program pays and on what terms. Abaka publishes its process but not its contract, so the last part of this page covers what to ask before you sign.
What is Abaka AI, and what does it sell?
Abaka AI sells data to companies that build AI models. Its homepage leads with "Human Intelligence Data for Frontier AI" and lists off-the-shelf datasets across text, audio, image, video, 3D, reasoning and agent tasks, plus data collection, annotation by "1M vertically specialized annotators across 50+ countries" and its own annotation software, the Abaka Forge Platform (Abaka AI).
Its datasets page lists ready-made sets such as a repo patch dataset for coding, Lean4 math formalization, stock videos with dense captions, talking-head videos and a browser recording dataset for agent training (Abaka AI datasets). Its newer RL environments product offers "Expert-built replicas of the enterprise apps companies run on": simulated mail, calendar, chat and document tools where AI agents practice office work (Abaka AI).
Unusually for this market, Abaka publishes starting prices. Its pricing page lists stock images from $0.01 an image, stock videos from $0.1 a video and Lean4 math formalization from $15 each (Abaka AI pricing). Those are prices Abaka charges its own customers. In our reading they matter to an owner for one reason: ordinary stock images and clips resell for cents, so the value in a company archive sits in what is rare, such as real work records with their history, not in volume alone.
Encord also sells annotation software and pays for data: it says it pays by the recorded hour for real work, and our Encord review covers what it asks a workplace to capture.
Source: Abaka AI: homepage
Abaka AI funding and ownership: what is on record?
In May 2026 Yahoo Finance reported that "Abaka.AI announced $8 million in funding to expand its enterprise data platform", quoting Yunfei Zhao as "co-founder and chief operating officer of Abaka.AI". Part of the money was to support the 2077AI Foundation, "an open-source initiative co-founded by Abaka.AI" (Yahoo Finance). The report did not name the investors.
An older record adds detail. A Forbes profile of co-founder Jack Lin, last updated in May 2024, says he "dropped out of his Ph.D. studies in computer science at Zhejiang University to cofound Abaka AI", that the startup launched in 2022, and that "Abaka AI has raised $20 million in total funding", with investors including Spinel Capital and Future Value Capital (Forbes). The profile lists him in Hangzhou, China.
The two figures do not add up to a clear total: $20 million by 2024 and an $8 million round in 2026 could both be true, but neither source gives a running figure. What the record does show is a founder based in Hangzhou in 2024 and a company that now hires in Silicon Valley: every role on its careers page, including a "Data Acquisition Engineer", is in Palo Alto, California (Abaka AI careers). For a seller, that matters in one practical way. Ask which Abaka entity signs your agreement and under which country's law.
Source: Yahoo Finance: Abaka.AI secures $8M to expand AI data infrastructure (May 2026)
Does Abaka AI pay experts and annotators?
Yes, but it publishes no pay rates for them. Abaka's annotation page claims "500,046 Registered experts" across 108 countries and recruits specialists such as physicians, attorneys, quants and software engineers for annotation, RLHF and model evaluation. It closes with an invitation: "Are you an expert? Join our network" (Abaka AI).
Abaka describes five checks before anyone touches client data: "application, an AI interview reviewed by a human, government-ID and location checks, a domain exam graded by senior reviewers, and continuous re-verification through completed work." That is in line with other expert networks. The hourly figures on its pricing page are what clients pay Abaka, not what experts earn, so ask for the rate in writing before you accept a project.
If you want paid expert work, compare Abaka with networks that publish rates; our directory of AI data companies lists them. If you are a company with records, the expert network is not your route. Project Reborn is.
Does Abaka AI buy company data? How Project Reborn works
Yes. Abaka's Project Reborn page says: "We license private codebases and company operating data to train and evaluate AI models." It names two lines, private codebases at "Up to $1M per repo" and company operating data at "Up to $5M per company", and tells owners of a qualifying repo, "Your repo matches what AI labs are buying" (Abaka AI Project Reborn). In other words, Abaka licenses what it takes on to AI labs, which it does not name.
For code, the best fit is a repo with 500 or more pull requests, three or more contributors, handwritten code from 2023 or earlier and full commit history intact. For company data, Abaka wants "The docs, chats, tickets and SOPs your teams produce every day" from teams of 10 or more employees across several departments, and says a company can be live or wound down.
The process has five steps: a two-minute form, a human reply within 48 hours, a metadata-only scan that runs on your own machine, "A concrete offer in writing within 14 days of your submission", and payment in two parts, where "The second follows once quality checks are complete." Abaka says valuation is free, with "No setup or listing fees, ever", and that an NDA is signed before anything sensitive changes hands.
| Question | Abaka AI Project Reborn (checked 18 Nov 2026) |
|---|---|
| What it buys | Private codebases, and company operating data such as docs, chats, tickets and SOPs |
| Published ceiling | Up to $1M per private codebase; up to $5M per company for operating data |
| Best fit, code | 500+ pull requests, 3+ contributors, handwritten code from 2023 or earlier, full commit history |
| Best fit, company data | Teams of 10 or more employees, several departments and roles, live or wound down |
| Before signing | A metadata-only scan on your machine; an NDA before sensitive material moves |
| Timing | Human reply within 48 hours; written offer within 14 days of submission |
| Payment | Two parts; the second after quality checks are complete |
| Cost to the seller | No upfront, setup or listing fees |
| Stated use | Training and evaluating AI models |
| End buyers | AI labs; none named |
Source: Abaka AI: Project Reborn
What Project Reborn does not publish
Abaka publishes more of its process than most buyers. The gaps are in the contract, and they are the parts that decide what you are paid.
- No sample agreement. The page sets out steps, not terms. Ask for the full license before the metadata scan, so you can read the payment trigger, the warranty and the scope while every option is still open.
- Ceilings, not prices. "Up to $1M" and "up to $5M" are maximums. Your figure arrives in the written offer, after the scan.
- The second payment depends on quality checks. Ask what the checks measure, who runs them, how long they take and what happens to the first payment if a check fails.
- Exclusivity is not stated. Ask whether the license is exclusive and for how long, because an exclusive deal stops you licensing the same code or records elsewhere.
- The end buyers are unnamed. Abaka says the data is "Used only to train and evaluate AI models", but the labs that receive it are not listed. Ask whether Abaka may sublicense, and to whom.
- Introducers are paid. Abaka runs a program that invites investors, advisors, agencies and former founders to "introduce a company that signs, and earn 10% of the deal". If someone outside Abaka brought you the offer, ask whether Abaka pays them. It does not make the offer worse, but you should know who is paid by whom.
Source: Abaka AI: Project Reborn
Got an offer from Abaka AI?
Start with the four checks we apply to every buyer, set against Abaka's published terms.
Our guide to what to check before you accept an AI data offer covers how to verify the sender and what a written offer must state, and the free data offer checker scores an offer you already hold. To compare Abaka's ceilings with what others pay for code and records, see selling source code to AI companies and company data buyers compared. Our 10-point checklist for vetting an AI data buyer goes further.
- Confirm who signs and who pays: the company itself, or an unnamed lab behind it. Get the paying entity named in the agreement. Abaka is the counterparty you deal with, but it licenses on to AI labs it does not name. Get the Abaka entity that pays you named with its country, and ask whether any lab receives your data directly.
- Get the price, the payment trigger and the payment date in writing. An estimator figure or an "up to" range is not an offer. Abaka promises a written offer within 14 days. Check that it states the total, the split between the two payments, the quality checks that release the second one and the date each is due.
- Read the license: exclusive or not, how long it lasts, whether the buyer may resell or build products on your records. The page limits use to training and evaluating AI models. Confirm that in the contract, along with exclusivity, term, territory and whether Abaka may resell or sublicense.
- Ask what is removed before delivery (names, customers, payroll, credentials, privileged legal material) and whether you approve a sample first. The metadata-only scan reads metadata, not content. Ask who removes names, customer details, credentials and secrets from the full set, and whether you approve a sample before delivery.
Next step
If Abaka has sent you a figure for a codebase or 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.
Common questions
Is Abaka AI legit, and does it pay companies for their data?
Yes on both counts. Abaka AI is an operating AI data company that sells datasets, annotation and model evaluation, hires in Palo Alto and announced $8 million in funding in May 2026. Its Project Reborn program pays companies to license private codebases and company operating data for AI training, with a written offer within 14 days. It does not publish a sample contract or name the AI labs that buy.
How much does Abaka AI pay for an old codebase?
Abaka advertises up to $1M per private codebase, which is a ceiling, not a price. The figure for your repo comes in a written offer within 14 days of submitting, after a metadata-only scan on your machine. Repos fit best with 500 or more pull requests, three or more contributors, handwritten code from 2023 or earlier and full commit history.
What company data does Abaka AI buy?
Abaka buys company operating data: the documents, chats, tickets and standard operating procedures teams produce every day. It says the best fit is a company with 10 or more employees across several departments, live or wound down, and it advertises up to $5M per company. It licenses the data on to AI labs for training and evaluating models.
Does Abaka AI charge sellers anything?
No, by its own account. Abaka says valuation is free and there are no upfront costs, setup fees or listing fees. It pays in two parts, with the second part released once quality checks are complete. Read how those quality checks work before you sign, because they decide when the second part arrives and how large it is.
Does my code leave my machine before I sign with Abaka AI?
Not according to Abaka. Before signing you run a local package that reads metadata only and writes a report you can read first, and Abaka says your code never leaves your machine before signing. It also offers an in-person review on your laptop in the Bay Area, and signs an NDA before anything sensitive changes hands.
How much funding has Abaka AI raised?
Yahoo Finance reported in May 2026 that Abaka AI announced $8 million in funding for its enterprise data platform and the 2077AI Foundation. A Forbes profile of co-founder Jack Lin, last updated in May 2024, said Abaka had raised $20 million in total from investors including Spinel Capital and Future Value Capital. Neither source gives a current running total.
Can I work for Abaka AI as an expert or annotator?
Yes. Abaka recruits domain experts such as physicians, attorneys, quants and engineers for annotation, RLHF and model evaluation, and its annotation page invites experts to join its network. Applicants pass five checks, including an AI interview reviewed by a human, ID and location checks and a domain exam. Abaka does not publish what it pays experts.
Which AI labs receive the data Abaka AI licenses?
Abaka does not say. It states that it licenses codebases and company data to train and evaluate AI models, and that qualifying repos match what AI labs are buying, but it names no lab. Ask in writing whether Abaka may sublicense your records, to whom, and whether you will be told which lab receives them before delivery.
Sources and further reading
- Abaka AI: Project Reborn
- Abaka AI: homepage
- Abaka AI: off-the-shelf datasets
- Abaka AI: pricing
- Abaka AI: data annotation and expert network
- Abaka AI: RL environments
- Abaka AI: careers
- Yahoo Finance: Abaka.AI secures $8M to expand AI data infrastructure (May 2026)
- Forbes: profile of Jack Lin, founder of Abaka AI (May 2024)
