DataPacts

Mechanize: RL Coding Environment Builder Profile

Checked

At a glance

Mechanize, Inc. is a US company, announced on April 17, 2025, that builds reinforcement learning (RL) environments and evals for frontier coding agents and sells them to AI developers as training material.

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

Mechanize's arithmetic: 500,000 tokens per run at USD 15 per million output tokens with group size 64 gives about USD 480 per run; five reuses give USD 2,400; about USD 500 is the minimum efficient build spend per task.

Its own pages state one funding round, dated April 24, 2026, and an argument of August 18, 2025 that building one RL task for less than about USD 500 wastes compute. No acquisition, license or customer contract involving Mechanize has been announced by Mechanize or a counterparty as of November 9, 2026.

Every figure on this page is Mechanize's own statement with its date, read in the context of AI training data licensing, where RL environments are one form of data acquired for AI training. DataPacts states no transaction that a party has not announced.

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

What Mechanize Builds and Sells

Mechanize describes its work in one line on its home page, checked 2026-11-09: "We build environments and evals for frontier coding agents." An environment in this sense is the task, the tools and the grader that a model trains against, and an eval is a task set used to score a model.

The home page names coding agents as the only target, which places Mechanize among builders of coding environments rather than computer-use or enterprise-replica builders.

Mechanize states in its announcement of April 17, 2025 that its mission is the full automation of the economy. The aim is the company's own statement; the announcement gives no date or measure for reaching it, and this page offers no assessment of it.

What Mechanize sells beyond that one line is not on record. No customer is named in the sources checked, no terms of sale or license are published, and no Mechanize program for licensing company data or paying contributors per task appears among the programs tracked as of 2026-11-09. How environments are built, graded and bought in general is explained under RL environments, which this profile does not repeat.

Mechanize Company Identity

Mechanize's legal name is Mechanize, Inc., as its own announcement gives it, and the company is based in the United States. The table lists each identity attribute with its source, as of 2026-11-09.

Attribute Value Source
Legal name Mechanize, Inc. mechanize.work/announcing-mechanize-inc
Website https://www.mechanize.work mechanize.work
Registry number none found on primary pages, checked 2026-11-09 n/a
Country United States mechanize.work/announcing-mechanize-inc
City not stated in the sources checked n/a
Founded 2025; announced 2025-04-17 mechanize.work/announcing-mechanize-inc
Type RL environment builder for frontier coding agents mechanize.work
Team around 35 people (Mechanize's claim, checked 2026-11-09) mechanize.work/what-working-here-is-like
Current chief executive not named in the sources checked n/a

Three attributes are not public in the sources checked: a state registry number, a city and the names of current officers. Each is added when a Mechanize page or a filing states it.

Who Announced Mechanize, and Who Leads It on the Record

Mechanize was announced on April 17, 2025 by Matthew Barnett, Tamay Besiroglu and Ege Erdil, in the company's own post "Announcing Mechanize, Inc." The same three names sign Mechanize's essay of August 18, 2025 on the cost of RL tasks, in the order Erdil, Barnett, Besiroglu, so they are also the authors of the company's published economics.

The sources checked give no titles for the three and name no current chief executive. DataPacts does not take roles or prior employers from profiles, directories or press coverage, so this page names no officer of Mechanize. The page is updated when Mechanize or a filing names one.

Mechanize Funding, as Mechanize States It

Mechanize states that it raised USD 9.1 million at a USD 500 million post-money valuation, in a press release dated April 24, 2026. The release names three investors: Marco Mascorro, Adam D'Angelo and Devendra Chaplot.

This page calls it a round without a type label, because no round type is confirmed in the sources checked. The table gives the round as stated, as of 2026-11-09.

Date Amount Post-money valuation Investors named Stated by Source
2026-04-24 USD 9,100,000 USD 500,000,000 Marco Mascorro, Adam D'Angelo, Devendra Chaplot Mechanize mechanize.work/mechanize-raises-9-1m

Mechanize's own statement; no filing in the record confirms it.

No filing confirms the round, so the amount and the valuation remain Mechanize's statement of 2026-04-24, and no other amount is stated by the company or a filing.

Funding is what investors paid for shares in the company, not a price a lab pays for an environment, and it says nothing about revenue. Rounds across companies are compared under funding and valuations of AI data companies.

Mechanize's Published Cost of One RL Task

Mechanize argued on August 18, 2025 that spending less than about USD 500 on building one RL task wastes compute, and projected that the minimum would rise to a few thousand dollars within a year.

The essay, signed by Erdil, Barnett and Besiroglu, starts from the compute a lab spends training on one task: a projected 500,000 tokens per run (current frontier tasks "around 100,000", in Mechanize's words, re-checked 2026-11-16) at USD 15 per million output tokens (the Grok 4 API price), times a group size of 64, gives about USD 480 per run, and five reuses give USD 2,400 over the task's life.

Each input is Mechanize's assumption. The table sets out the figures from the essay of 2025-08-18, as of 2026-11-09.

Item Value Stated by Date Source
Tokens per training run 500,000 (projection; about 100,000 currently) Mechanize 2025-08-18 mechanize.work/blog/cheap-rl-tasks-will-waste-compute
Price per million output tokens USD 15 Mechanize 2025-08-18 mechanize.work/blog/cheap-rl-tasks-will-waste-compute
Group size 64 Mechanize 2025-08-18 mechanize.work/blog/cheap-rl-tasks-will-waste-compute
Compute per run USD 480 Mechanize 2025-08-18 mechanize.work/blog/cheap-rl-tasks-will-waste-compute
Reuses of one task 5 Mechanize 2025-08-18 mechanize.work/blog/cheap-rl-tasks-will-waste-compute
Compute per task over its life USD 2,400 Mechanize 2025-08-18 mechanize.work/blog/cheap-rl-tasks-will-waste-compute
Minimum efficient spend per task about USD 500, projected a few thousand within a year Mechanize 2025-08-18 mechanize.work/blog/cheap-rl-tasks-will-waste-compute
For comparison: price range per task USD 200 to 2,000 Epoch AI interviewees 2026-01-12 (publication) epoch.ai/gradient-updates/state-of-rl-envs

Assumptions and projections are Mechanize's; the last row is a different party's statement and is not combined with Mechanize's figures.

Epoch AI, in an article published on January 12, 2026, cites the same lifetime compute figure from Mechanize and reports that its interviewees put a task at USD 200 to 2,000.

The two disagree at the low end: a task priced at the bottom of that range costs less than Mechanize's stated floor, so on Mechanize's reasoning it wastes compute. DataPacts shows both and does not average them; they sit beside the other price claims for RL environment work on the hub.

The essay is an argument about how much building effort a task deserves, not a price list, and it does not state what Mechanize charges. A seller of tasks can use it as one input next to the checks labs run before buying, which the guide to selling environments covers.

Mechanize Transactions and Status: What the Primary Record Shows

No acquisition, license or other transaction involving Mechanize has been announced by Mechanize or by a counterparty in the sources DataPacts checks, as of November 9, 2026, so this page states none and gives no value.

Coverage that rests on neither a party's statement nor a filing is not used on this site, however often it is repeated.

The tracker of AI data licensing deals lists a transaction only when a party or a filing confirms it, and none of its records names Mechanize as of 2026-11-09. The company list for RL environment builders shows the same result for acquisitions.

Mechanize's careers page states a team of around 35 people, checked 2026-11-09, which is the company's own claim. This page is updated when Mechanize, a counterparty or a filing publishes a change to the company's ownership, staff or status.

Working With Mechanize: Benefits, Risks, Downsides and Alternatives

For a data holder or an environment builder, Mechanize is a builder that publishes its economics but no terms: its essay states how much building spend it thinks a task deserves, while no contract, license or data program of Mechanize is on record.

Benefits

  • Published reasoning on the cost of a task, with every input stated (essay of 2025-08-18).
  • A dated round with three named investors, stated by the company on 2026-04-24.
  • A single stated focus: environments and evals for frontier coding agents.

Risks

  • No published Mechanize terms on who owns deliverables or how supplied data is used. Proximal, another coding builder, states in its MSA that it owns the deliverables (s.8.1, checked 2026-11-09), so ownership cannot be assumed either way.
  • No registry number found on primary pages, checked 2026-11-09.
  • No officer named on the record.

Downsides

  • No program for licensing company data or codebases and no contributor bounties on record as of 2026-11-09.
  • No named customer, so no reference transaction exists to compare against.

Alternatives

  • Other builders on the company list below, each with its own terms.
  • Marketplaces where the vendor sets the price: DataVendor states a 20 % fee on earnings at withdrawal and a USD 1,000 minimum withdrawal, checked 2026-11-09.
  • Contributor task platforms such as Shipd, operated by Datacurve AI, Inc., whose Contributor Terms (s.3.1) assign all work product to Shipd, checked 2026-11-09.
  • Direct sale: the routes to sell an RL environment or task set compare build contracts, marketplaces and bounties.
  • Not licensing or contributing at all, which is a valid outcome.

Mechanize's figures are its own statements and can change, so its current pages are the reference before relying on any of them.

Mechanize Among RL Environment Builders

Mechanize is one of several companies that build RL environments for coding agents, listed with every other builder on DataPacts' company list. The list of RL environment companies places it among custom environment builders for labs, beside nine others. Related pages:

  1. Other builder profiles: Fleet AI, Proximal, a coding builder with a published MSA, and Datacurve, which runs the Shipd task platform.
  2. The environment type and its prices: RL environments for coding agents and the public price points for RL environments.
  3. How tasks are scored and what holders of code can license: verifiers and graders and licensing source code to AI companies.
  4. The full decision, including not selling: whether to sell or license data at all.

More About Mechanize

Frequently Asked Questions About Mechanize

Who founded Mechanize?

Mechanize was announced on April 17, 2025 by Matthew Barnett, Tamay Besiroglu and Ege Erdil. That is how the company's own announcement presents them. It gives no titles, and this page adds none from other sources.

Who is the CEO of Mechanize?

The sources checked name no current chief executive of Mechanize, as of November 9, 2026. DataPacts does not take the role from profiles or press coverage and adds a name once Mechanize or a filing states one.

Is Mechanize a real company?

Yes, on the record: Mechanize, Inc. has its own website, a dated announcement of April 17, 2025 and a funding release of April 24, 2026. No registry number was found on its primary pages, checked 2026-11-09.

Where is Mechanize based?

Mechanize is based in the United States; the sources checked give no city. A city is added once a Mechanize page or a filing states an address for the company.

How much funding has Mechanize raised?

Mechanize states that it raised USD 9.1 million at a USD 500 million post-money valuation, in a release dated April 24, 2026. No other amount is stated by the company or a filing as of 2026-11-09.

Does Mechanize buy company data?

No Mechanize program for licensing company data is on record as of November 9, 2026. Mechanize builds and sells environments and evals for coding agents, and holders of codebases find buyer routes on the source-code page listed above.