# RL environment companies and deals tracker (2026): who builds them and who pays for company data

> Who builds RL environments for AI labs: 18 companies with funding, domains and whether they pay for company data, plus deals and what environments cost.

Canonical: https://origindatapartners.com/guides/rl-environment-companies
By: Origin Data Partners
Status: Published
Published: 2026-11-04
Updated: 2026-11-04
Sources checked: 2026-11-04

## Quick answer

RL environment companies build simulated workplaces where AI agents practise tasks and are scored: replicas of CRMs, ticketing tools, code repositories and finance workflows, filled with realistic data. The largest suppliers are human-data companies that added environments (Mercor, Surge AI, Scale AI, Handshake AI, Turing, micro1) and specialists such as Prime Intellect, Halluminate, Deeptune (acquired by Mercor), HUD, Plato and Patronus AI. Several pay companies for operational data to make environments realistic: Handshake AI says its partnerships typically run $100,000 to $4 million, and micro1 committed $1 billion over 12 months. Contracts with labs often run six to seven figures per quarter, according to Epoch AI.

## Key points

- Most environment builders sell to a handful of frontier labs; Anthropic alone discussed spending more than $1 billion on environments over a year.
- Builders pay for company data because realism is the bottleneck: a Slack-grade replica can cost about $300,000 to build.
- Finance, enterprise apps, coding and computer use are where most new environments are being built.

## RL environment companies compared

Funding and scale are as reported by the linked source. "Pays for company data" means the company or a named report says it buys or licenses operational data from businesses.

| Company | What it builds | Domains | Funding or scale | Pays for company data? | Source |
| --- | --- | --- | --- | --- | --- |
| Mercor | Environments, tasks and verifiers for frontier labs; acquired Deeptune | Coding, healthcare, law, finance, enterprise apps | $350M Series C at $10B (Oct 2025) | Yes: licenses and pays for company data | [Mercor](https://www.mercor.com/data) |
| micro1 | Environments built from de-identified company data | Enterprise operations and workflows | $1B data commitment financed by Citi and Hercules Capital (Oct 2026) | Yes: $100K to $2M per approved data package | [micro1](https://www.micro1.ai/blog/micro1-commits-1b-to-enterprise-data-acquisition) |
| Handshake AI | Environments built with employer data partnerships | Finance (investment banking), enterprise workflows | AI-training revenue near $1B annualised (Apr 2026) | Yes: partnerships typically $100K to $4M | [Dealroom](https://dealroom.co/news/127345-handshakes-arr-crosses-1b-as-ai-training-revenue-surges/) |
| Surge AI | Internal RL-environments organisation | Enterprise IT, legal and other professional work | Reported $1.2B revenue in 2024 | Yes: buys data to build "RL gyms" | [MIT Technology Review](https://www.technologyreview.com/2026/10/13/1145755/ai-labs-buying-data-from-dead-startups-to-train-models/) |
| Scale AI | Scale RL Environments for training and evaluation | Tool use, computer use, coding | Nearly half of new data projects involve environments | Not stated | [Scale AI](https://scale.com/blog/rl-environments) |
| Turing | Coding data and environments for labs | Coding | $111M Series E at $2.2B (Mar 2025) | Yes: bought 5 to 10 failed-startup code bases at tens of thousands each | [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/) |
| Prime Intellect | Open Environments Hub, RL framework and compute | Open community environments: coding, math, tool use | $130M Series A at $1B; $100M revenue run rate (Jul 2026) | Not stated | [TechCrunch](https://techcrunch.com/2026/07/08/prime-intellect-raises-130m-series-a-to-help-enterprises-build-their-own-ai-agents/) |
| Halluminate | Benchmarks and environments for knowledge work | Finance and consulting | $30M Series A, $38.5M total (Oct 2026) | Yes, per an investor: licenses operational data, up to six or seven figures | [Fortune](https://fortune.com/2026/10/01/halluminate-raises-30-million-series-a-oakhc-ft/) |
| Deeptune | High-fidelity replicas of workplace software | Slack, Salesforce, ticketing, finance | $43M Series A (Mar 2026); acquired by Mercor (Jul 2026) | Not stated | [Fortune](https://fortune.com/2026/03/19/andreessen-horowitz-ai-startups-deeptune-series-a/) |
| HUD | Platform and vendor marketplace for building and selling environments | Coding, computer use, browser | $16M Series A; 2,500+ environments (Jun 2026) | Not stated | [HUD](https://www.hud.ai/blog/announcing-16m-series-a) |
| Plato | Simulated workplaces for real jobs | Customer support, people ops, finance, accounting, healthcare admin | $16M Series A, $19M total (Oct 2026) | Pays professionals to recreate their jobs | [PR Newswire](https://www.prnewswire.com/news-releases/plato-raises-16-million-to-build-the-training-grounds-for-ais-next-leap-into-real-work-302921832.html) |
| Patronus AI | "Digital world models" to train and stress-test agents | Software engineering, finance | $50M Series B, $70M total (Jun 2026) | Not stated | [TechCrunch](https://techcrunch.com/2026/06/25/patronus-ai-lands-50m-to-build-digital-worlds-that-stress-test-ai-agents/) |
| Mechanize | Software-engineering environments | Coding | $9.1M at $500M post-money (Apr 2026); Google talent deal (Sept 2026) | Not stated | [Business Insider](https://www.businessinsider.com/google-completes-deal-for-ai-agents-startup-mechanize-2026-9) |
| Bespoke Labs | Platform that generates and runs environments | Agent post-training | $31.75M Series A (Jul 2026) | Not stated | [SiliconANGLE](https://siliconangle.com/2026/07/06/ai-post-training-startup-bespoke-labs-raises-40m-funding/) |
| Ooak Data | Anonymised digital twins of a company's tools | Gmail, Slack, Notion, Jira, Drive, SharePoint | Above seven figures in revenue (YC, Aug 2026) | Yes: $4M+ paid out to companies, per its CEO | [Y Combinator](https://www.ycombinator.com/launches/Sih-ooak-data-turn-internal-company-data-into-a-training-ground-where-ai-agents-can-be-trained-on-real-work) |
| Abaka AI | Full-stack simulated SaaS apps | Enterprise messaging, e-commerce, recruiting | Built a recruiting environment from de-identified internal data | Uses de-identified internal data | [Abaka AI](https://www.abaka.ai/blog/rl-environments-enterprise-ai-agents) |
| Polymath | Long-horizon, multi-tool software environments | Software engineering | More than $8M raised | Not stated | [Polymath](https://jobs.ashbyhq.com/polymath/44812b4a-b41c-4288-8232-e0131da6eefe) |
| Veris AI | Simulation environments for enterprise agents | Enterprise agents | $8.5M seed (Jun 2025) | Not stated | [Veris AI](https://veris.ai/blog/introducing-veris-ai-a-new-way-to-train-enterprise-ai-agents-through-simulated-experience) |


## RL environment deals tracker

Funding rounds, acquisitions and data commitments with a named source, newest first.

| Date | Company | Deal | Value | Source |
| --- | --- | --- | --- | --- |
| Nov 2026 | Handshake AI | Partnership with Fireworks AI on enterprise RL environments | Undisclosed | [Handshake](https://joinhandshake.com/research/ai/fireworks-handshake-rl-environments/) |
| Oct 2026 | Plato | Series A led by Nava Ventures | $16M | [PR Newswire](https://www.prnewswire.com/news-releases/plato-raises-16-million-to-build-the-training-grounds-for-ais-next-leap-into-real-work-302921832.html) |
| Oct 2026 | Handshake AI | Employer Data Partnerships launched to buy company data for environments | $100K to $4M per partnership (typical) | [Handshake](https://joinhandshake.com/blog/hai/handshake-ai-employer-data-program/) |
| Oct 2026 | micro1 | Commitment to acquire and license company operational data over 12 months | $1B | [micro1](https://www.micro1.ai/blog/micro1-commits-1b-to-enterprise-data-acquisition) |
| Oct 2026 | Halluminate | Series A led by Oak HC/FT | $30M | [Fortune](https://fortune.com/2026/10/01/halluminate-raises-30-million-series-a-oakhc-ft/) |
| Sept 2026 | Mechanize | Google talent deal and technology licence completed | Talks reported at $1.5B+; final terms undisclosed | [Business Insider](https://www.businessinsider.com/google-completes-deal-for-ai-agents-startup-mechanize-2026-9) |
| Sept 2026 | micro1 | Series B at a $4B valuation | $4B valuation | [micro1](https://www.micro1.ai/blog/q3) |
| Aug 2026 | micro1 | Data-licensing commitments to seed environments over 11 days | $20M+ | [CEO on LinkedIn](https://www.linkedin.com/posts/aliansarinik_in-the-last-11-days-weve-committed-more-activity-7495381329388163072-y7gr) |
| Jul 2026 | Mercor | Acquisition of Deeptune | Undisclosed | [Mercor](https://www.mercor.com/blog/mercor-to-acquire-deeptune/) |
| Jul 2026 | Prime Intellect | Series A led by Radical Ventures at a $1B valuation | $130M | [TechCrunch](https://techcrunch.com/2026/07/08/prime-intellect-raises-130m-series-a-to-help-enterprises-build-their-own-ai-agents/) |
| Jul 2026 | Bespoke Labs | Series A led by Wing VC | $31.75M | [SiliconANGLE](https://siliconangle.com/2026/07/06/ai-post-training-startup-bespoke-labs-raises-40m-funding/) |
| Jun 2026 | HUD | Series A led by Standard Capital | $16M | [HUD](https://www.hud.ai/blog/announcing-16m-series-a) |
| Jun 2026 | Patronus AI | Series B led by Greenfield Partners | $50M | [TechCrunch](https://techcrunch.com/2026/06/25/patronus-ai-lands-50m-to-build-digital-worlds-that-stress-test-ai-agents/) |
| Apr 2026 | Mechanize | Funding at a $500M post-money valuation | $9.1M | [Mechanize](https://www.mechanize.work/mechanize-raises-9-1m/) |
| Mar 2026 | Deeptune | Series A led by Andreessen Horowitz | $43M | [Fortune](https://fortune.com/2026/03/19/andreessen-horowitz-ai-startups-deeptune-series-a/) |
| Oct 2025 | Mercor | Series C at a $10B valuation | $350M | [TechCrunch](https://techcrunch.com/2025/10/27/mercor-quintuples-valuation-to-10b-with-350m-series-c/) |
| Jun 2025 | Veris AI | Seed round led by Decibel and Acrew | $8.5M | [Veris AI](https://veris.ai/blog/introducing-veris-ai-a-new-way-to-train-enterprise-ai-agents-through-simulated-experience) |
| Mar 2025 | Turing | Series E at a $2.2B valuation | $111M | [TechCrunch](https://techcrunch.com/2025/03/06/turing-a-key-coding-provider-for-openai-and-other-llm-producers-raises-111m-at-a-2-2b-valuation/) |


## What RL environments cost

Epoch AI interviewed 18 people across environment startups and labs for a [January 2026 FAQ](https://epochai.substack.com/p/an-faq-on-reinforcement-learning). Contracts often run six to seven figures per quarter. A replica of a website costs about $20,000 to build, while a high-fidelity replica of a tool like Slack can cost about $300,000. Individual tasks typically cost $200 to $2,000 each, and exclusive deals cost four to five times more than non-exclusive ones. Spending on environments is still small next to compute, and labs are bringing more of the work in-house.

Labs are leaning in: Anthropic's leaders discussed spending more than $1 billion on environments over a year, [TechCrunch reported](https://techcrunch.com/2025/09/21/silicon-valley-bets-big-on-environments-to-train-ai-agents/) from The Information in September 2025, and nearly half of Scale AI's new data projects now involve environments, [it said](https://scale.com/blog/rl-environments) in February 2026.


## RL environments by domain

| Domain | What the environment contains | Builders in this tracker | Company data that fits |
| --- | --- | --- | --- |
| Coding and software engineering | Repositories with failing tests, CI, logs and issue trackers | Mechanize, Turing, Polymath, Scale AI, Prime Intellect | [Source code with history](https://origindatapartners.com/guides/sell-source-code-ai-training) |
| Computer use and browser agents | Replicas of websites and desktop apps | HUD, Scale AI | Product usage sessions, see [AI usage logs](https://origindatapartners.com/guides/license-ai-product-usage-data) |
| Finance and consulting | Deal rooms, models, filings and banking workflows | Halluminate, Handshake AI, Patronus AI | [Deal and portfolio records](https://origindatapartners.com/guides/license-deal-data-investment-firms) |
| Enterprise apps | CRM, ticketing, chat and email replicas filled with records | Deeptune, Mercor, micro1, Ooak Data, Abaka AI | [Slack and email archives](https://origindatapartners.com/guides/sell-slack-email-archives-ai-training), [support tickets](https://origindatapartners.com/blog/license-support-tickets-for-ai) |
| Customer support and operations | Ticket queues, knowledge bases and back-office tools | Plato, micro1 | Tickets with resolutions, procedures |
| Healthcare and law | Clinical admin and legal drafting workflows | Mercor, Plato, Surge AI | [Patient records](https://origindatapartners.com/guides/license-patient-records-ai-training), law-firm templates |


## What this means for a company with data

Environment builders compete on realism, and realism comes from real records: messy, linked and with outcomes attached. That is why several now pay companies directly. The buyers with published ranges in this tracker pay $100,000 to $4 million per partnership, and wind-down sales of smaller startups' archives run from about $10,000 to $100,000. See [what RL environments are](https://origindatapartners.com/guides/rl-environments-company-data) for which records make good environment data.


## Who buys company data, and who helps you sell it

Buyers, intermediaries and what deals have paid:

- [Who buys company data for AI training?](https://origindatapartners.com/guides/who-buys-company-data-ai-training)
- [Brokers, marketplaces and advisors compared](https://origindatapartners.com/guides/company-data-brokers-compared)
- [AI data buyers compared](https://origindatapartners.com/guides/company-data-buyers-compared)
- [Can your company license its operational data?](https://origindatapartners.com/guides/company-data-licensing-programs)
- [What AI buyers pay: 2026 study](https://origindatapartners.com/guides/ai-company-data-buyer-programs-study)
- [AI data licensing deals tracker](https://origindatapartners.com/guides/ai-data-licensing-deals)
- [How much do AI companies pay for data?](https://origindatapartners.com/guides/company-data-value)
- [RL environments and why labs buy company data](https://origindatapartners.com/guides/rl-environments-company-data)
- [Selling company data to hedge funds](https://origindatapartners.com/guides/sell-company-data-hedge-funds)
- [Where to sell data to hedge funds, compared](https://origindatapartners.com/guides/sell-data-to-hedge-funds-platforms)
- [SimpleClosure review and alternatives](https://origindatapartners.com/guides/simpleclosure-alternatives-selling-startup-data)


## Next step

If your company holds tickets, chat, CRM history, code or finance workflows, Origin can tell you which environment builders would want them and what they are likely to pay.

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](https://origindatapartners.com/referral-disclosure)). [See what your data is worth](https://origindatapartners.com/check-fit).


## Questions and answers

### Which companies build RL environments?

Human-data companies that added environments, such as Mercor, Surge AI, Scale AI, Handshake AI, Turing and micro1, and specialists such as Prime Intellect, Halluminate, HUD, Plato, Patronus AI, Bespoke Labs, Polymath and Ooak Data. Deeptune was acquired by Mercor in July 2026.

### What are the biggest RL environment startups?

By reported funding, Prime Intellect raised $130 million at a $1 billion valuation in July 2026, Patronus AI $50 million, Deeptune $43 million before Mercor acquired it, and Halluminate $30 million. Mercor, valued at $10 billion, calls itself the largest RL environments company.

### Do RL environment companies buy company data?

Several do. Handshake AI says its partnerships typically run $100,000 to $4 million, micro1 committed $1 billion over 12 months and pays $100,000 to $2 million per approved package, Surge AI buys data for its RL gyms, and Ooak Data's CEO says it has paid companies more than $4 million.

### How much does an RL environment cost?

Epoch AI reported that a website replica costs about $20,000 and a Slack-grade replica about $300,000 to build, tasks cost $200 to $2,000 each, and lab contracts often run six to seven figures per quarter. Exclusive deals cost four to five times more.

### What are RL environments for finance?

Simulated deal rooms, models, filings and banking workflows where agents practise analyst tasks. Halluminate, Handshake AI and Patronus AI build them, and finance firms' historical records are the realistic data they need.

### What is an RL gym?

Another name for an RL environment: a simulated setting where an AI agent practises tasks and an automatic grader scores it. The term comes from earlier reinforcement learning toolkits and is still used by builders that sell "RL gyms" to labs.

## Sources

- [Epoch AI: An FAQ on reinforcement learning environments (Jan 2026)](https://epochai.substack.com/p/an-faq-on-reinforcement-learning)
- [TechCrunch: Silicon Valley bets big on environments to train AI agents (Sept 2025)](https://techcrunch.com/2025/09/21/silicon-valley-bets-big-on-environments-to-train-ai-agents/)
- [Scale AI: RL environments (Feb 2026)](https://scale.com/blog/rl-environments)
- [micro1: $1B commitment to enterprise data acquisition (Oct 2026)](https://www.micro1.ai/blog/micro1-commits-1b-to-enterprise-data-acquisition)
- [Handshake: employer data partnerships (Oct 2026)](https://joinhandshake.com/blog/hai/handshake-ai-employer-data-program/)
- [MIT Technology Review: AI labs buying data from dead startups (Oct 2026)](https://www.technologyreview.com/2026/10/13/1145755/ai-labs-buying-data-from-dead-startups-to-train-models/)
- [Fortune: Halluminate raises $30M (Oct 2026)](https://fortune.com/2026/10/01/halluminate-raises-30-million-series-a-oakhc-ft/)
- [Fortune: Deeptune Series A (Mar 2026)](https://fortune.com/2026/03/19/andreessen-horowitz-ai-startups-deeptune-series-a/)
- [TechCrunch: Prime Intellect raises $130M (Jul 2026)](https://techcrunch.com/2026/07/08/prime-intellect-raises-130m-series-a-to-help-enterprises-build-their-own-ai-agents/)
- [Mercor: acquiring Deeptune (Jul 2026)](https://www.mercor.com/blog/mercor-to-acquire-deeptune/)
- [HUD: $16M Series A (Jun 2026)](https://www.hud.ai/blog/announcing-16m-series-a)
- [Y Combinator: Ooak Data launch (Aug 2026)](https://www.ycombinator.com/launches/Sih-ooak-data-turn-internal-company-data-into-a-training-ground-where-ai-agents-can-be-trained-on-real-work)
- [Business Insider: Google completes Mechanize deal (Sept 2026)](https://www.businessinsider.com/google-completes-deal-for-ai-agents-startup-mechanize-2026-9)

## Related reading

- [What are RL environments? How AI labs build them and why they pay for company data](https://origindatapartners.com/guides/rl-environments-company-data)
- [License workflow data for AI agents: a company owner’s guide](https://origindatapartners.com/guides/license-workflow-data-for-ai-agents)
- [AI data licensing deals tracker: who paid what, 2023 to 2026](https://origindatapartners.com/guides/ai-data-licensing-deals)
- [Can an AI startup license its usage data and agent logs?](https://origindatapartners.com/guides/license-ai-product-usage-data)