Introducing Pangaea and Emet: AI that knows the rules of global work

Today we’re launching Emet, our new AI lab, and Pangaea, our family of models for global work.

Why? AI may have accessed the entire internet, but it’s never run payroll or a business.

The best AI model fully solves only 49% of real payroll tasks. On global work questions, like notice periods or who can legally start a job, the best model gets 75% of the facts right, even with search. Close enough to look right. Wrong enough to break a paycheck or a contract.

Deel’s payroll engines bring decades of experience running payroll, HR, and compliance, now in 150+ countries. That’s expertise the internet doesn’t have. So we built Emet to put it to work.

Emet is our AI lab for model evaluation and training, specializing in how businesses run: payroll, HR, compliance, legal, and finance. Its experts are practicing specialists from Deel’s own teams and its global network, and Deel finds, verifies, and pays them. They define what real business tasks look like and what it takes to get them done right. Emet turns that into public benchmarks and training data, starting with payroll and global work. Emet is open to every model builder.

Pangaea is the first proof of what Emet can do. It’s Deel’s family of LLMs for global work: hiring, paying, and managing people across 150+ countries. The models are built on leading open-weight models and trained on Deel’s knowledge base and expert-built data from Emet, never on personal or customer data. Pangaea is rolling out as the main model behind AI at Deel.

Emet: where experts define right

Fast answers only matter if they’re right. Emet’s specialists turn what they know into two things: public benchmarks that show where AI falls short, and training data that helps it get the job done.

The benchmarks are called EmBench. Every model, ours included, is scored against expert answers. The first two:

• EmBench Payroll. 295 real payroll tasks across 8 countries, built with 29 in-country payroll specialists. The best model meets 90% of their criteria but fully solves only 49% of the tasks.

• EmBench Global Work. 601 questions on working across borders, from notice periods to who can legally start work, scored fact by fact. The best model, with search, gets 75% of the facts right.

Why this matters to you. Your team may already use AI for HR and payroll questions. EmBench shows where you can trust it, and where you can’t yet. Pangaea is on the same public leaderboards, so anyone can check our results.

Pangaea: faster answers, right inside Deel

As Pangaea becomes the main model behind AI at Deel, here’s what changes for your team.

• Faster answers on global work. Ask Deel about notice periods, leave rules, hiring in a new country, or payroll, and you’ll get a full answer in about 8 seconds. Leading AI models with search took up to 28 seconds in our testing.

• Answers held to an expert standard. In-country payroll, compliance, and HR specialists define what a right answer looks like. Pangaea is tested against that standard, and the results are public.

• Your data stays yours. Pangaea is never trained on customer payroll or HR records.

• AI stays included. Pangaea answers at a fraction of the cost of leading AI models. That helps us keep AI included in your Deel plan, with no charge per question.

How you use Deel stays the same. It will just get faster.

Meet the first two Pangaea models

Pangaea K1: Global Work. K1 knows the rules of working across borders: notice periods, leave rules, hiring in a new country. It’s trained on Deel’s knowledge base of 43,500+ pages, so it answers from what it already knows instead of searching. On EmBench Global Work, K1 comes within five points of the top model with search, answering from memory, up to 3x faster.

Pangaea P1: Payroll. P1 knows payroll across countries, where one missed detail means a wrong paycheck. It was trained on payroll data built by Emet’s in-country specialists. On the 110 hardest tasks in EmBench Payroll, that training took P1 from 55% to 61% of criteria passed, at about 40x lower cost than the top model.

Proven on Deel’s own AI first

AI at Deel is tested on our own work first. Emet’s specialists check what it does, catch what it gets wrong, and turn those misses into training data for the next version. Emet measured every step as training on Deel’s knowledge base took K1 from 19% to 70% of facts right, answering from memory.

So the AI your team uses in Deel is held to the standard of experts who do this work every day.

What comes next

Pangaea will keep getting faster and more accurate. Emet will add benchmarks across more of how businesses run, starting with payment reconciliation exceptions and sales tasks, and will build environments where AI agents practice real business tasks from start to finish. The goal stays the same: AI that gets your work right, so your people can focus on the decisions that matter.