Explore and compare models

Why it matters

Each agent role needs a model chosen from evidence, and this exercise walks the Foundry tools that produce that evidence before a deployment is committed.

What to practise

Read model cards and benchmarks in the Microsoft Foundry model catalog, then use the model leaderboard’s trade-off charts to set quality against cost, throughput and safety for two models. Deploy both, compare them side by side in the playground, and run an evaluation whose dataset Foundry generates synthetically, scored by AI-assisted evaluators. Watch the results analysis cluster failures by cause. The exercise notes that some of its features are in preview.

How it fits

Use it with 1.1.11: compare context, endpoints, agent support and cost per role rather than picking the top-ranked model. For 3.2, look at how the evaluation is built: the synthetic dataset is a starting point for 3.2.8’s regression cases, and the judge-based evaluators are what 3.2.7 says to calibrate before trusting their scores.