EmBench Payment Reconciliation
Real reconciliation tasks, scored by finance experts.
Benchmarks for business operations
Introducing EmBench
EmBench shows how AI models handle the work that runs a business, starting with payroll and global work. Every model, ours included, is scored against expert answers.
Live
EmBench
Payroll
Real payroll questions across 8 countries, built with in-country payroll experts.
Measuring AI models’ ability to give complete, correct payroll answers where national and local rules differ.
Claude Opus 5.5
89.7%Claude Sonnet 5.5
82.5%Gemini 3.8 Flash
78.9%Pangaea
73.6%GPT-6 Luna
70.7%
Live
EmBench
Global Work
601 questions on hiring, payroll, and compliance across borders, scored fact by fact.
Measuring AI models’ ability to answer cross-border work questions, from notice periods to who can legally start.
DeepSeek V4.1 Flash
75%Claude Opus 5.5
74%GLM 5.3 Flash
72%Claude Sonnet 5.5
71%Pangaea
70%
Coming soon to EmBench
Coming soon Coming soon EmBench SDR
Real prospecting and outreach tasks, scored by sales experts.
By the people who do the work. For the people who do the work
Tell us where your model needs to improve. We’ll show you how it performs today, and build what it needs to get better.
Get notified when we launch a new benchmark
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.
Emet is Deel’s AI lab. Deel runs payroll, HR, and compliance for 40,000+ companies in 150+ countries, so our first benchmarks come from work Deel does every day. Deel’s network also lets Emet find, verify, and pay experts compliantly around the world.
EmBench is Emet’s family of benchmarks for business operations. It tests how AI models handle the work that runs a business, starting with payroll and the rules of working across borders. Every model, ours included, is scored against expert answers. Each answer is scored on how complete and correct it is, not just whether it sounds right. Each benchmark’s page has its full details.
Tasks start from the questions and jobs Deel’s teams handle every day, with no personal or client data. In-country experts build and correct the answer keys. A judge model scores each answer against what a complete answer must include, checked against expert reviews. Each benchmark’s report has the full method.
Talk to us about the areas your model needs to get better at. We can evaluate it and build training data with the same experts and process.
No. Test questions and answers stay private, so scores reflect real skill, not memorization.
