Build an eval set
Real tasks from your teams, in both languages.
Start · (02) · Model selection
We test open-weight and commercial models on your own tasks, in Arabic and English, and pick per agent for accuracy, cost, speed and where the model is allowed to run.
Timeframe
2 to 4 weeks
You keep
An evaluation set and a model plan you can re-run.
(02)
Model selection
The problem
Public leaderboards measure English trivia, not your Arabic forms. One model for everything is either too expensive for simple tasks or too weak for hard ones.
Real tasks from your teams, in both languages.
Open-weight and commercial, side by side.
Accuracy, cost per task, latency, hosting.
The right model for each job, swappable later.
What you get
Timeframe · 2 to 4 weeks
Where it fits
Yes. We include open-weight models you can host yourself.
You re-run the eval set and switch. Nothing else changes.
A one-hour meeting with our team, on a real example from your organization.