DataPacts

BenchFlow: RL Environment Builder Profile

Checked

At a glance

BenchFlow is a US company that describes itself as "a frontier environment lab for AI agents" and ships benchmarks, RL environments, an open-source runtime and PostTrain Arena, where developers contribute environments, checked 2026-11-09. BenchFlow states no compensation for a contributed environment and publishes no legal entity name, and no customer, licensing deal or court case involving it is on record.

Independent reference. Every buyer is covered on the same terms, every figure carries its source and check date, and no buyer pays to be listed or placed higher.

Request an independent introduction

Tell us what data your company holds. We reply with the buyers that fit and what their terms leave out. Your details reach a buyer only after you agree to a named introduction.

IndependentYour details stay with us until you agreeNot selling is a valid outcome

Every BenchFlow figure on this page is BenchFlow's own statement with its check date, and each other company's figure is that company's claim, read in the context of AI training data licensing, where an RL environment is one form of material used to train models.

No database, aggregator or social-profile figure is used, so the legal name, founding year and funding stay blank where BenchFlow publishes none.

Each value names its source and check date, and how DataPacts sources and dates every figure is set out in the methodology.

What BenchFlow Builds: Benchmarks, Environments and a Runtime

BenchFlow ships six named products on its own site: SkillsBench, ClawsBench, Robo Use, PostTrain Arena, FrontierPhysics and an Apache-2.0 runtime, checked 2026-11-09.

The table lists what BenchFlow states about each product (as of 2026-11-09).

Product What BenchFlow states Source
SkillsBench 87 tasks across 8 domains (BenchFlow's claim) benchflow.ai
ClawsBench named product, no further detail recorded benchflow.ai
Robo Use 341 tasks in 14 suites (BenchFlow's claim) benchflow.ai
PostTrain Arena takes contributed environments, post-trains models on them and scores them benchflow.ai
FrontierPhysics named product, no further detail recorded benchflow.ai
Runtime open source, Apache-2.0 github.com/benchflow-ai/benchflow

The two task counts are BenchFlow's claims: SkillsBench at 87 tasks across 8 domains and Robo Use at 341 tasks in 14 suites. DataPacts records no count for the other products, and it does not describe them beyond their names.

An RL environment is the task, the tools and the checker that an agent is scored against. For what RL environments are and who buys them, the hub page holds the definition and the lab-side evidence.

BenchFlow sits on both sides of that market: it publishes benchmarks of its own and takes in environments from outside contributors.

The distinction matters for a builder. SkillsBench and Robo Use are task sets that BenchFlow itself defines and counts, while PostTrain Arena scores environments that other developers build, so the arena is the product through which outside work reaches BenchFlow.

BenchFlow Company Identity: What Is and Is Not Published

BenchFlow publishes no legal entity name: its site footer reads "© 2026 BenchFlow", and no terms or privacy page names a contracting company, checked 2026-11-09.

The table compares what BenchFlow's own pages state with what they leave out (as of 2026-11-09).

Attribute Value Source
Brand BenchFlow benchflow.ai
Website benchflow.ai benchflow.ai
Legal name not published; footer "© 2026 BenchFlow" benchflow.ai
Registry and Form D none found n/a
Country United States benchflow.ai
City not recorded n/a
Founded not stated in the sources recorded n/a
Type RL environment builder, platform segment benchflow.ai
Funding not stated by the company or a filing n/a

No Form D was found for the company, and no registry entry is recorded. The founding year and any funding round are not stated by BenchFlow or by a filing in the sources DataPacts records, so the page leaves them blank and repeats no database or summary figure.

Rounds that companies do state are set side by side in the page on funding and valuations of AI data companies.

PostTrain Arena: BenchFlow's Terms for Contributed Environments

BenchFlow invites anyone to contribute an RL environment to PostTrain Arena, with the line "Contribute an environment. We post-train and score what generalizes.", and states no compensation for it, checked 2026-11-09.

The table sets out BenchFlow's own statements for each term of the arena (as of 2026-11-09). BenchFlow's own statements; DataPacts has no relationship with BenchFlow.

Term BenchFlow's statement Checked Source
What it takes in RL environments 2026-11-09 benchflow.ai
Eligibility open 2026-11-09 benchflow.ai
Process submit to PostTrain Arena 2026-11-09 benchflow.ai
Pay no compensation stated 2026-11-09 benchflow.ai
Referral fee none stated 2026-11-09 benchflow.ai
License not stated on benchflow.ai; posttrainarena repository AGPL-3.0 2026-11-09 benchflow.ai, github.com/benchflow-ai/posttrainarena

For a contributor, the rows read as follows. Eligibility is open, and the process is a submission to the arena. BenchFlow post-trains models on submitted environments and scores which ones generalize.

No payment, revenue share, prize or referral fee is stated, and the license for a contribution is not stated on benchflow.ai, while the arena's GitHub repository carries AGPL-3.0.

The same row sits in the cross-program comparison of program terms compared across buyers and builders, where it can be read against the terms of paid programs.

BenchFlow's Repository Licenses and What They Leave Open

BenchFlow publishes its runtime under Apache-2.0 and its PostTrain Arena repository under AGPL-3.0, two different open-source licenses in the same organisation, checked 2026-11-09.

The site does not state, for a contributed environment, which license covers it, whether the contributor keeps ownership, whether BenchFlow may reuse models post-trained on it, how attribution works or how a contribution is taken down.

The parent list already records that open hubs state no assignment of the contributor's work, and BenchFlow's arena is one of the two it names.

A contributor therefore cannot read the answers to these questions from the site alone, so each one is a point to settle in writing before a submission, and the repository license describes the code in that repository, not the terms for an environment submitted to the arena.

How license terms travel when code is licensed for AI training is covered on its own page, open-source license contamination, which applies the same questions to a licensed codebase.

BenchFlow on the Record: Customers, Deals and Cases

No customer of BenchFlow is named in the sources DataPacts checks, and no licensing deal, acquisition or court case involving BenchFlow is on record as of November 9, 2026.

No BenchFlow program that pays for company data or for contributed environments is on record either. The only program recorded for BenchFlow is PostTrain Arena, which takes contributed environments without a stated payment.

The tracker of AI data licensing deals lists a transaction only with a primary source, and none names BenchFlow. The page is updated when BenchFlow, a counterparty or a filing publishes a change.

Until then, the absence of a named customer is a statement about the sources checked on 2026-11-09, not a statement about what BenchFlow does privately, and a reader weighing a contribution should treat it that way.

Contributing to BenchFlow: Benefits, Risks, Downsides and Alternatives

For a developer who has built an RL environment, PostTrain Arena is a place to publish it and have models post-trained and scored on it, not a sale: BenchFlow states no compensation, checked 2026-11-09.

Benefits

  • Public scoring of whether an environment generalizes, as BenchFlow describes the arena.
  • Open eligibility: BenchFlow states no restriction on who may contribute.
  • An open repository, so the contributor can read the arena's code under its AGPL-3.0 license.

Risks

  • No published legal entity to contract with, because BenchFlow names none.
  • The license for a contribution is not stated on the site.
  • No stated terms on ownership, reuse or takedown of a contributed environment.
  • The arena repository's AGPL-3.0 license differs from the runtime's Apache-2.0.

Downsides

  • No payment is stated for a contributed environment.
  • No named customer or deal is on record, so the environment has no stated buyer.

Alternatives

  • Prime Intellect's Open Access environment bounties, stated "in the $100-500 range", Prime Intellect's claim, checked 2026-11-09.
  • General Reasoning's OpenReward, which also states no payment to environment creators, checked 2026-11-09.
  • DataVendor, run by HUD, where the vendor sets the price and DataVendor states a 20 % fee on earnings at withdrawal and a USD 1,000 minimum withdrawal, checked 2026-11-09.
  • To sell an RL environment or task set to an AI lab directly.
  • Keeping the environment private and not contributing at all, a valid outcome.

BenchFlow's statements are its own and can change, so the current page and the repository license are worth checking before a contribution.

The weighing is simple to state: the arena offers public scoring and an open codebase, and it offers no stated payment, no named buyer and no stated terms on what happens to the contributor's rights.

Whether that trade suits a builder depends on whether the environment has a better use elsewhere, and a builder with a paying route can compare it against the alternatives above.

BenchFlow Among Open Environment Hubs

BenchFlow sits with the platforms, hubs and marketplaces on DataPacts' company list, beside Eigent AI, General Reasoning, HUD and Prime Intellect.

Each of these takes in or lists environments from outside builders, and they differ in what they state about payment: Prime Intellect states bounties, General Reasoning and BenchFlow state none, and HUD's DataVendor states a fee at withdrawal.

The full list of RL environment companies places each one by segment, and the related pages below cover the questions a contributor meets next.

More About BenchFlow

Frequently Asked Questions About BenchFlow

What is BenchFlow?

BenchFlow is a US environment and benchmark lab for AI agents. It ships SkillsBench, ClawsBench, Robo Use, PostTrain Arena, FrontierPhysics and a runtime, as BenchFlow describes itself, checked 2026-11-09.

Does BenchFlow pay for RL environments?

PostTrain Arena states no compensation for contributed environments, checked 2026-11-09. No program of BenchFlow that pays for environments or company data is on record.

Is BenchFlow open source?

Its runtime repository is licensed Apache-2.0 and its PostTrain Arena repository AGPL-3.0, per GitHub, checked 2026-11-09. The site states no license for a contributed environment.

What is SkillsBench?

SkillsBench is BenchFlow's benchmark, which BenchFlow states has 87 tasks across 8 domains, checked 2026-11-09. The figure is BenchFlow's own claim.

Has BenchFlow raised funding?

BenchFlow and the filings in the sources checked state no round. DataPacts does not repeat database figures for BenchFlow's funding.

Does BenchFlow buy company data?

No BenchFlow program for acquiring company data is on record as of 2026-11-09. PostTrain Arena takes contributed environments without a stated payment.