About SimReal
We build the worlds where intelligence learns to meet ours.
Our mission
We build worlds that answer back.
Chess, Go and code gave intelligence a clear answer: a winner, a failing test, another try. Most economic work gives no such signal. SimReal rebuilds that work as environments where an agent’s actions have consequences, and grades it by what actually happens, not by another model’s opinion.
How we work
Four commitments behind every score.
- 01
Mechanistic, not learned
The reward is the mechanism itself. There is no judge model to game.
- 02
A human denominator
Where one exists, we score against what people actually achieved.
- 03
Failures on the record
We attack every grader before release and keep a record of what gets through. On the Puzzle Benchmark, 400 cheating attempts were run and none did.
- 04
Independent of model companies
No AI model company owns any part of SimReal. A measurement is worth what its independence is worth.
Built by a team with experience from
Company facts
- Founded
- 2026
- Company focus
- RL environments, verifiers and benchmarks for AI agents doing economic work
- Products
- Five products on one engine: SimReal-MLBench, Puzzle Benchmark, MathmoBench, Future Prediction Bench and SWE-Forward
- Domains
- AI research, reasoning, optimisation, forecasting and software
- Who we serve
- Frontier AI labs first, then companies deploying AI agents
- Access
- Benchmarks, audit records and data specifications are shared with partners on request
- Model-company shareholders
- None
- External capital
- None
- Contact
- business@simreal.co
Available on request
- SimReal-MLBench: sixty research tasks graded by the platforms that hold the answers Request access
- Six task families · 18 scenario templates Request a demo
- Expert trajectory data specification Request the specification