Environments for Frontier AI

Eval Labs researches benchmarks and evaluates frontier models, then builds reinforcement learning environments, at scale, to push what those models can do.

Built by researchers and engineers from Meta Superintelligence Labs.

We've worked with teams at

  • Anthropic
  • UC Berkeley
  • Stanford
  • OpenAI
  • TBD Lab
  • xAI
  • Google DeepMind

What we do

Find the limits of frontier models. Then push past them.

Our work spans two areas that feed each other.

  • Benchmark research and model evaluations

    We design benchmarks and run evaluations that show where frontier models fall short, and why.

  • Environments

    We build RL environments and graders, then scale them across tasks and difficulty to push models past the frontier.

How it works

Your agent says the task is done. Our graders check.

We grade on outcomes, not on a model's account of what it did.

  1. Measure

    Benchmark research and evaluations pinpoint where frontier models fail.

  2. Build

    We build environments and graders aimed squarely at those gaps.

  3. Scale

    We scale environments across tasks and difficulty to push the frontier forward.

Work with us

Every engagement starts with a call.

AI labs

Tell us what you're training and where your models fall short, and we'll walk you through how we'd build for it: the evaluations, the environments and how they're graded.

Get in touch

Partners

We collaborate with domain experts and specialist firms. If you'd like to partner with us, we'd like to hear from you.

Become a partner