AISandboxEnvironments

What is paid to build RL environments and UI gyms

4 of 12 builders, buyers and platforms for RL environments and UI gyms publish what is paid on the pages we read, read 8 October 2026.

4 of 12 payers Payers that publish a figure of the payers read from their own pages on 8 Oct 2026
8 of 12 Unit of the figure: not stated the commonest of 5 answers
9 of 12 Low end of the stated figure: not stated the commonest of 3 answers
9 of 12 High end of the stated figure: not stated the commonest of 3 answers
11 of 12 What is built: stated payers whose pages state it, with the quote below
15 Pages examined 23 named sources; every answer is dated and hashed

Publishes a figure

  1. no figure published8 payers
  2. publishes a figure4 payers
12 payers counted, one answer apiece; conflicting pages from one payer are counted as "varies".
Counts
RowValue
no figure published8 payers
publishes a figure4 payers

Unit of the figure

  1. not stated8 payers
  2. a contract or deal1 payer
  3. per hour of building1 payer
  4. per task1 payer
  5. annual salary1 payer
Across 12 payers, a single answer each.
Counts
RowValue
not stated8 payers
a contract or deal1 payer
per hour of building1 payer
per task1 payer
annual salary1 payer

Low end of the stated figure

  1. not stated9 payers
  2. a figure in US dollars2 payers
  3. varies1 payer
Across 12 payers, a single answer each.
Counts
RowValue
not stated9 payers
a figure in US dollars2 payers
varies1 payer

High end of the stated figure

  1. not stated9 payers
  2. a figure in US dollars2 payers
  3. varies1 payer
Across 12 payers, a single answer each.
Counts
RowValue
not stated9 payers
a figure in US dollars2 payers
varies1 payer

What is built

  1. stated11 payers
  2. not stated1 payer
Across 12 payers, a single answer each.
Counts
RowValue
stated11 payers
not stated1 payer

How every payer answered

A payer per row. Every answer opens onto the line it comes from, quoted exactly, with the page it sits in and the day of the read. "Not stated" marks a question the page leaves unanswered.

PayerPublishes a figureUnit of the figureLow end of the stated figureHigh end of the stated figureWhat is built
Anthropic and OpenAI
publishes a figure

In 2025, Anthropic planned to spend over $1 billion on reinforcement learning environments over the next year, and OpenAI signed multiple seven-figure environment contracts in the same window. Prolific: 2026 guide to enterprise RL environments

a contract or deal

OpenAI signed multiple seven-figure environment contracts in the same window Prolific: 2026 guide to enterprise RL environments

not stated

Not addressed in Prolific: 2026 guide to enterprise RL environments

not stated

Not addressed in Prolific: 2026 guide to enterprise RL environments

not stated

Not addressed in Prolific: 2026 guide to enterprise RL environments

Careerflow
publishes a figure

$50 USD per task, paid once the task is accepted and approved in review. Careerflow: AI/ML Software Engineer, Task Creator (RL Environments)

per task

$50 USD per task, paid once the task is accepted and approved in review. Careerflow: AI/ML Software Engineer, Task Creator (RL Environments)

50 USD

$50 USD per task, paid once the task is accepted and approved in review. Careerflow: AI/ML Software Engineer, Task Creator (RL Environments)

50 USD

$50 USD per task, paid once the task is accepted and approved in review. Careerflow: AI/ML Software Engineer, Task Creator (RL Environments)

Each task is a small scenario running on a Linux desktop…

Each task is a small scenario running on a Linux desktop — real files, real applications, real documents — plus a program that automatically checks whatever the agent produced. Careerflow: AI/ML Software Engineer, Task Creator (RL Environments)

Collinear
no figure published

Every simulation lab is a self-contained world where your agent operates, complete with the users, tools, data, and tasks it will face in production. Collinear home page

not stated

Not addressed in Collinear home page

not stated

Not addressed in Collinear home page

not stated

Not addressed in Collinear home page

From GUIs and APIs to MCP, CLI, GitHub, Jira, ServiceNow, and Salesforce

From GUIs and APIs to MCP, CLI, GitHub, Jira, ServiceNow, and Salesforce Collinear home page

Deeptune
no figure published

We build training gyms for frontier research labs. Deeptune home page

not stated

Not addressed in Deeptune home page

not stated

Not addressed in Deeptune home page

not stated

Not addressed in Deeptune home page

We've built 100s of gyms that mimic popular software like Slack and Salesforce.

We've built 100s of gyms that mimic popular software like Slack and Salesforce. Deeptune home page

Halluminate
no figure published

Halluminate is a data research lab building benchmarks and RL environments to push the frontier of AI in knowledge work. Halluminate home page

Halluminate is building realistic sandboxes and datasets to train better computer/browser use AI. YC launch: Halluminate, data and sandboxes to train computer use AI

not stated

Not addressed in Halluminate home page

Not addressed in YC launch: Halluminate, data and sandboxes to train computer use AI

not stated

Not addressed in Halluminate home page

Not addressed in YC launch: Halluminate, data and sandboxes to train computer use AI

not stated

Not addressed in Halluminate home page

Not addressed in YC launch: Halluminate, data and sandboxes to train computer use AI

It draws on anonymized data from real private transactions, runs inside dynamic desktop…

It draws on anonymized data from real private transactions, runs inside dynamic desktop environments, and spans trajectories that reach hundreds of steps. Halluminate home page

Halluminate is building realistic sandboxes and datasets to train better computer/browser use AI. YC launch: Halluminate, data and sandboxes to train computer use AI

HUD
publishes a figure

San Francisco (and other major US cities): $135,000 - $230,000 HUD: Research Engineer (YC jobs)

annual salary

Actual offers are adjusted for experience and location, but our base salary bands are HUD: Research Engineer (YC jobs)

135000 USD

San Francisco (and other major US cities): $135,000 - $230,000 HUD: Research Engineer (YC jobs)

230000 USD

San Francisco (and other major US cities): $135,000 - $230,000 HUD: Research Engineer (YC jobs)

HUD's mission is to build reliable, fair and open infrastructure for AI data.

HUD's mission is to build reliable, fair and open infrastructure for AI data. HUD: Research Engineer (YC jobs)

Keymakr
no figure published

We can help build, manage, and scale high-fidelity reinforcement learning environments (RL-Gyms) where your agents can learn to reason, plan, and recover from errors. Keymakr: virtual environments and RL gyms

not stated

Not addressed in Keymakr: virtual environments and RL gyms

not stated

Not addressed in Keymakr: virtual environments and RL gyms

not stated

Not addressed in Keymakr: virtual environments and RL gyms

Environments that replicate your target deployment - a Unix shell, a Salesforce CRM…

Environments that replicate your target deployment - a Unix shell, a Salesforce CRM sandbox, or a live e-commerce site replica. Keymakr: virtual environments and RL gyms

Mechanize
no figure published

We build environments and evals for frontier coding agents. Mechanize home page

not stated

Not addressed in Mechanize home page

not stated

Not addressed in Mechanize home page

not stated

Not addressed in Mechanize home page

In these environments, models carry out software engineering work such as building a…

In these environments, models carry out software engineering work such as building a feature, deploying an application, or debugging an issue in an unfamiliar codebase. Mechanize home page

Mercor
publishes a figure

New opportunity Early applicant $75-$100 per hour Mercor: Experienced Software Engineers (RL environments)

1 hired this month Early applicant $25-$30 per hour Mercor: Technical Reviewer, RL Environment Terminal Benchmarking (India)

per hour of building

New opportunity Early applicant $75-$100 per hour Mercor: Experienced Software Engineers (RL environments)

1 hired this month Early applicant $25-$30 per hour Mercor: Technical Reviewer, RL Environment Terminal Benchmarking (India)

varies by document

New opportunity Early applicant $75-$100 per hour Mercor: Experienced Software Engineers (RL environments)

1 hired this month Early applicant $25-$30 per hour Mercor: Technical Reviewer, RL Environment Terminal Benchmarking (India)

varies by document

New opportunity Early applicant $75-$100 per hour Mercor: Experienced Software Engineers (RL environments)

1 hired this month Early applicant $25-$30 per hour Mercor: Technical Reviewer, RL Environment Terminal Benchmarking (India)

Your role will focus on designing, developing, and maintaining high performance RL…

Your role will focus on designing, developing, and maintaining high performance RL environments. Mercor: Experienced Software Engineers (RL environments)

To review, validate, and improve reinforcement learning environment benchmarking pipelines Mercor: Technical Reviewer, RL Environment Terminal Benchmarking (India)

Prime Intellect
no figure published

Not addressed in Prime Intellect Environments Hub listing

If you are interested in RL environments that don't have a bounty figure listed yet, just ask and we'll figure something out based on the difficulty scale we're using. Prime Intellect: Environments Hub launch

not stated

Not addressed in Prime Intellect Environments Hub listing

Not addressed in Prime Intellect: Environments Hub launch

not stated

Not addressed in Prime Intellect Environments Hub listing

Not addressed in Prime Intellect: Environments Hub launch

not stated

Not addressed in Prime Intellect Environments Hub listing

Not addressed in Prime Intellect: Environments Hub launch

DeepSWE long-horizon software engineering tasks from Harbor Hub.

DeepSWE long-horizon software engineering tasks from Harbor Hub. Prime Intellect Environments Hub listing

Prime Sandboxes are fully capable Linux virtual machines built for agentic training, with pricing designed for thousands of concurrent instances. Prime Intellect: Environments Hub launch

Surge AI
no figure published

As an RL Environments Architect, you’ll design, instrument, and govern the simulated worlds where agents learn Surge AI: RL Environments Architect

not stated

Not addressed in Surge AI: RL Environments Architect

not stated

Not addressed in Surge AI: RL Environments Architect

not stated

Not addressed in Surge AI: RL Environments Architect

Architect a modular environment framework with clear APIs, curriculum scaffolds, and…

Architect a modular environment framework with clear APIs, curriculum scaffolds, and configurable reward/termination schemas Surge AI: RL Environments Architect

Veris AI
no figure published

A simulated copy of your systems, data, APIs, and users, where AI gets tested, trained, and built against before it ever touches production. Veris AI home page

not stated

Not addressed in Veris AI home page

not stated

Not addressed in Veris AI home page

not stated

Not addressed in Veris AI home page

Give coding agents a stateful simulation of your dependencies, so integration tests run…

Give coding agents a stateful simulation of your dependencies, so integration tests run and issues get fixed before the PR opens Veris AI home page

The source sentence behind every answer (75)

Documents we read but did not count

6 named documents were read and left out because they are not what this board covers; each reason is below. They are counted, never answered.

Method

Each payer's own postings, pricing and buyer pages were read through estate-fetch and stored whole (sha256 on every row).

For every page, five questions were answered from its own words: publishes a figure, unit of the figure, low end of the stated figure, high end of the stated figure, what is built.

Every stated answer carries the sentence it rests on, checked against the stored text; 'not stated' means the page read does not address the question.

A payer with several pages reads 'varies' where they disagree.

12 payers, 15 pages.

2 named documents have not been read and are excluded from the counts.

Source list (23)
Take this figure with you

4 of 12 builders, buyers and platforms for RL environments and UI gyms publish what is paid on the pages we read, read 8 October 2026.

Publishes a figure, by payer: no figure published 8 of 12, publishes a figure 4 of 12 (12 payers, read 8 October 2026).

Unit of the figure, by payer: not stated 8 of 12, a contract or deal 1 of 12, per hour of building 1 of 12, per task 1 of 12, annual salary 1 of 12 (12 payers, read 8 October 2026).

Low end of the stated figure, by payer: not stated 9 of 12, a figure in US dollars 2 of 12, varies 1 of 12 (12 payers, read 8 October 2026).

High end of the stated figure, by payer: not stated 9 of 12, a figure in US dollars 2 of 12, varies 1 of 12 (12 payers, read 8 October 2026).

What is built, by payer: stated 11 of 12, not stated 1 of 12 (12 payers, read 8 October 2026).

Citation: "What is paid to build RL environments and UI gyms", updated 2026-10-08, https://aisandboxenvironments.com/datasets/sandbox-environment-buyers-and-specs/.

Put this figure on your page (HTML only)
4 of 12 builders, buyers and platforms for RL environments and UI gyms publish what is paid on the pages we read, read 8 October 2026.
What is paid to build RL environments and UI gyms, aisandboxenvironments.com

Raw files: JSON, CSV, Markdown.

The data as a CSV (75 answers; derived 2026-10-08)

Describe my environment