Videos · 2:49 · 17 Nov 2026

What AI labs pay for RL environments and training tasks, and how to sell them (2026)

An RL environment is a sandbox in which an AI agent attempts a task with the tools it is given, while a verifier checks the result and turns it into a score that trains the model. Labs buy them by the task and by the environment. Here is what they pay, and what makes one worth buying.

Figures: aidata.broker/guides/rl-environments, each price with its published source.

Transcript

What the video says

0:00What do AI labs pay for RL environments and training tasks?

An RL environment is a sandbox in which an AI agent attempts a task with the tools it is given, while a verifier checks the result and turns it into a score that trains the model. Labs buy them by the task and by the environment. Here is what they pay, and what makes one worth buying.

0:24By the task, and by the environment

Builders and lab researchers interviewed by Epoch AI put a task with its grader at two hundred to two thousand dollars. Epoch, citing SemiAnalysis, puts a working replica of a website at about twenty thousand dollars, and a high-fidelity clone of an app like Slack at about three hundred thousand. And Scale AI wrote, as Fortune reported, that nearly half of its new data-training projects involve environments.

0:58Five things, in their own words

Buyers are consistent about what they pay for. A sound verifier: high reward must mean the task was actually solved, not hacked. Calibrated difficulty: tasks current models pass sometimes, not always or never. Realism: an environment that behaves like the real software. No contamination: a task that has appeared anywhere public is worth little. And expertise that is hard to replicate.

1:31What you hold is the material environments are built from

Most holders will never build an environment, and they do not need to. What they hold is the material environments are built from. A codebase, each change joined to its ticket and its tests, is a set of ready-made coding tasks. Company records become a simulated company for an agent to work in. And an expert's knowledge becomes the hardest tasks and the rubric that grades them.

2:00Only as licensable as what is inside it

An environment is only as licensable as everything inside it. Tasks written by contractors need a written assignment to whoever sells them. Records used to populate a simulated company need the same clearance as any sale of company data, de-identification included. And tasks built from someone else's material carry that material's restrictions.

2:26See what your tasks or records are worth, free

To see what your task sets, environments, code or records are worth, value them free at aidata.broker/value. One press introduces you to each buyer whose published rules take them, and you pay us nothing: the buyer pays us its referral fee. The full guide is on our site.

See what your company's data is worth, and which buyers take it.