Emet home

The AI lab for business operations

Benchmarks and training data from the experts who run global business every day.

Introducing EmBench

Benchmarks for business operations, built with the experts who do the work.

  • Live

    EmBench Global Work

    The rules of working across borders, scored fact by fact.

    1. DeepSeek V4.1 Flash

      75%
    2. Claude Opus 5.5

      74%
    3. GLM 5.3 Flash

      72%
    4. Claude Sonnet 5.5

      71%
    5. Pangaea

      70%
  • Live

    EmBench Payroll

    Real payroll questions across 8 countries, built with in-country experts.

    1. Claude Opus 5.5

      89.7%
    2. Claude Sonnet 5.5

      82.5%
    3. Gemini 3.8 Flash

      78.9%
    4. Pangaea

      73.6%
    5. GPT-6 Luna

      70.7%
  • Coming soon

    EmBench Payment Reconciliation

    Real reconciliation tasks, scored by finance experts.

  • Coming soon

    EmBench SDR

    Real prospecting and outreach tasks, scored by sales experts.

Benchmarks and expert data for business operations

Benchmarks

Real-world tests that show how your model performs on the work businesses actually do.

Talk to us

Expert training data

RL environments

Coming soon

Built by the people who run global work

Deel runs payroll, hiring, and compliance for 40,000+ companies in 150+ countries and we know what a right answer looks like. The knowledge that runs a business rarely lives on the public web. We built Emet to capture it, tested it on Deel’s AI first, and now open it to every model builder.

Deel knows how business runs

We see it at scale. Deel runs payroll, HR, and compliance for 40,000+ companies in 150+ countries.

We live it. 7,000+ people run Deel’s own finance, sales, and operations, on Deel.

Proven on Deel’s AI first. Emet helps build Pangaea, Deel’s family of LLMs, built on open-weight models.

We already have the experts. Deel finds, verifies, and pays specialists anywhere in the world.

That’s what Emet brings to every benchmark and dataset.

Quality you can check

  • Verified by experts

    Experts who do the work review every task before it reaches you.

    Join as an expert
  • Benchmarks first

    We publish the benchmark first, so you can see how frontier models perform before you commit. Then we build the data that improves your model.

  • Tested on our own AI

    Before working with any lab, we tested Emet on Deel’s own AI. It still runs there every day.

Frequently asked questions

Emet is Deel’s AI lab for model evaluation and training, specializing in how businesses run: payroll, HR, compliance, legal and finance. We build benchmarks and expert training data, and we’re building RL environments, to make AI better at the work that runs a business, starting with payroll and global work. Emet is open to every model builder.