Guides7 min read

RL Environment Companies in 2026: The Full Landscape

Troveo Team

Troveo

Reinforcement learning environments went from research plumbing to a funded vendor category in about a year. The reason is agents: once labs started training models to complete long-horizon tasks in software, they needed realistic places for those agents to practice, with verifiable rewards to learn from. Building a convincing replica of Salesforce, a hospital records system, or a full coding workflow is real work, and a wave of companies now sells exactly that.

Article banner reading RL Environment Companies, mapping the agent training market

This guide maps who builds RL environments in 2026, how the vendors differ, and the part of the stack that environments do not replace: the real-world data underneath.

What an RL environment actually is

An RL environment is a controlled world where an agent acts and gets scored. In practice that means three components: a simulated task space, such as a replica of enterprise software, a codebase, a browser, or an operating system; a goal specification, often written by domain experts; and a verifier that decides whether the agent succeeded, which is where most of the engineering difficulty lives. Weak verifiers get gamed, and reward hacking is the failure mode everyone in this market is quietly fighting. Our explainer on what RL environments are covers how they train an agent, the kinds labs use, and what they cost. Our guide to training data for AI agents covers what the environments are built from.

Three kinds of vendors sell them

The first group is the human data companies that added environments to existing RLHF and evaluation businesses. Scale AI sells simulated web apps, desktop VMs, and tool-based environments. Surge AI built an enterprise simulation suite. Mercor added environments alongside its expert networks, including through acquisitions. Our guide to Mercor alternatives covers that company's full trajectory.

Turing and Centific pitch environments as part of broader data services; we cover Centific's version in our guide to Centific alternatives. The pitch here is one vendor for data, evals, and environments together.

The second group is environment-native startups, founded for this market specifically. Mechanize builds high-fidelity coding environments for frontier labs. Fleet AI replicates enterprise software like CRMs and spreadsheets in high fidelity. HUD wraps real software as agent-callable tools in containers. Veris AI and Plato build simulated enterprise and web worlds. Bespoke Labs comes at it from the tooling side, with open-source libraries for curation and evaluation. These companies tend to be deeper on a specific environment type and faster-moving than the incumbents. If you're building on that side of the market, our guide to how to sell RL environments to AI labs covers who buys, what they pay, and the three ways to reach them.

The third group is open ecosystems and infrastructure. Prime Intellect hosts thousands of community-built environments with an open verifier library. General Reasoning runs a community hub with a unified API. Underneath all of it sit sandbox and compute providers like Modal and E2B, which run the containers everyone else's environments execute in.

The landscape

CompanyCategoryKnown for
Scale AIData giantSimulated web apps, desktop VMs and tool-based environments
Surge AIData giantEnterprise simulation suite
MercorData giantEnvironments alongside its expert networks
CentificData giantEnvironments as part of broader data services
TuringData giantCoding environments over real and private codebases
MechanizeEnvironment-nativeHigh-fidelity coding environments for frontier labs
Fleet AIEnvironment-nativeEnterprise software replicas
HUDEnvironment-nativeReal software wrapped as agent-callable tools
Veris AIEnvironment-nativeSimulated enterprise and web worlds
Bespoke LabsEnvironment-nativeOpen tooling for curation and evals
Prime IntellectOpen ecosystemCommunity hub of thousands of environments
Modal, E2BInfrastructureSandboxed compute the environments run on
The RL environment vendor landscape by category, 2026.

Which RL environment vendor fits which job

Most vendors go deep on one or two kinds of environment, so start from what your agent has to do.

  • Coding and software engineering agents. Mechanize builds high-fidelity coding environments for frontier labs, and Turing builds coding environments over real and private codebases.
  • Computer-use and browser agents. Scale AI sells simulated web apps, desktop VMs and tool-based environments. HUD wraps real software as tools an agent can call, and Veris AI and Plato build simulated web worlds.
  • Enterprise software and knowledge-work agents. Fleet AI replicates enterprise software like CRMs and spreadsheets, Surge AI built an enterprise simulation suite, and Veris AI and Plato build simulated enterprise worlds.
  • One vendor for data, evaluation and environments. Mercor sells environments alongside its expert networks, and Centific and Turing sell them as part of broader data services.
  • Open environments you run yourself. Prime Intellect hosts thousands of community-built environments with an open verifier library, General Reasoning runs a community hub with a unified API, and Modal and E2B supply the sandboxes they run in.

For narrower jobs like security auditing, report writing or data analysis, expect custom work. Bring the vendor your task list and ask which of their existing environments come closest.

Whichever vendor you pick, ask three things about reward hacking before you sign:

  1. How do you test a verifier against an agent that's trying to game it?
  2. Which failure cases have you caught, and what did you change?
  3. Can we run our own agent against a sample environment first?

How to buy in a market this young

Three cautions. First, definitions vary wildly: one vendor's "environment" is a full software replica with expert-written goals and verifiers, another's is a sandbox with a prompt. Ask exactly what ships. Second, verifier quality is the product. A beautiful simulation with a gameable reward teaches your agent to cheat; ask how verifiers are tested against reward hacking. Third, most environment work is still custom, so treat public offerings as starting points and scope carefully, the same discipline we recommend for choosing any data vendor.

Environments do not replace data

Here is the part of the stack the environment wave tends to obscure. An environment teaches an agent to act, but the model inside the agent still has to understand the world it is acting in, and that understanding comes from training data. World models learn physics and object behavior from real video. Agents that operate software learn interfaces from real demonstrations. Game-playing and embodied models learn from gameplay data, task demonstrations, and first-person footage of humans actually doing things.

Simulation and real-world data are complements: the environment provides practice and verification, the data provides the ground truth the simulation is built from and the model is grounded in. Labs that skip the second half get agents that ace the benchmark and fumble reality.

Where Troveo fits

Troveo does not build RL environments. Troveo licenses the real-world interaction data that agent and world-model teams train on: gameplay footage, task demonstrations, tool manipulation, first-person navigation, and real workflows, sourced from more than 7,000 rights holders and cleared for AI training with documentation per asset. If you are standing up an environment program, that data is what grounds the model before the environment sharpens it. Our guide to AI training data marketplaces covers how labs buy licensed data like this. Talk to us about what your agents need to learn from.

Frequently asked questions

What is an RL environment for AI training?
A controlled world where an agent acts and gets scored: a simulated task space such as replicated software or a codebase, goals usually written by domain experts, and a verifier that decides success. The verifier is the hard part; weak ones get gamed.
Who are the main RL environment companies in 2026?
Three groups: human data companies that added environments (Scale AI, Surge AI, Mercor, Turing, Centific), environment-native startups (Mechanize, Fleet AI, HUD, Veris AI, Plato, Bespoke Labs), and open ecosystems like Prime Intellect, with sandbox infrastructure from providers like Modal and E2B underneath.
What are the top alternatives to Scale AI for RL environments?
Surge AI and Mercor on the full-service side, Mechanize for frontier coding environments, Fleet AI and HUD for enterprise software environments, and Prime Intellect for the open-source route. Vendor neutrality is worth weighing here the same way it is in labeling, given Meta's stake in Scale.
Which RL environment companies also handle human evaluation and RLHF?
The data giants: Scale AI, Surge AI, Mercor, Turing, and Centific all sell environments alongside RLHF and human evaluation. Environment-native startups mostly do not; they partner or integrate instead.
What is RL Environments-as-a-Service?
A managed offering where the vendor designs, builds, and operates environments for you rather than shipping a product you run yourself. Centific coined the framing for its enterprise offering, and most custom environment engagements work this way in practice regardless of the label.
How much do RL environments cost?
Almost everything is custom-scoped, priced by environment complexity, expert involvement in goals and verifiers, and ongoing operation. Open-source hubs are free to start and cost engineering time instead. Treat any public pricing as a floor for negotiation.
Which RL environment companies build coding environments?
Mechanize builds high-fidelity coding environments for frontier labs, and Turing builds coding environments over real and private codebases.
How do you check an RL environment for reward hacking?
Ask the vendor how they test verifiers against an agent that’s trying to game them, which failure cases they’ve caught, and whether you can run your own agent against a sample before you sign.
What is the difference between an RL environment and training data?
Data teaches the model what the world is; an environment tests and sharpens what the model does. Real video, demonstrations, and interaction data build the model's understanding, then environments give it a place to practice acting on that understanding with verifiable feedback. Serious agent programs buy both.
Does Troveo build RL environments?
No. Troveo licenses the real-world interaction data that agent and world-model training runs on: gameplay, task demonstrations, tool manipulation, and first-person footage, rights-cleared with per-asset documentation. Environments are the practice field; Troveo supplies the ground truth.

Related articles

Back to Resources