Tutorial: let models debate, then measure them

Which model should get your hardest work? Don’t guess — make them compete. The Model Council has models argue a question in rounds; the Benchmark suite scores them on your actual tasks.

1. Convene a Council

Open Tools → Model Council:

Give it a question that matters (“How should we structure the sync engine?”), seat two or more models, and run. Each states a position, reads the others, and rebuts — you get the strongest version of every side, not one model’s first take.

2. Benchmark on YOUR tasks

Public leaderboards test someone else’s problems. Tools → Benchmark tests yours:

Add tasks (real prompts from your work), pick contenders, run. Results are scored side by side, so “which model writes our kind of code best” becomes a number instead of a vibe.

3. Act on the results

Winners earn assignments: set the strong model as an agent window’s model, point the Kanban Crew’s Judge at the most critical one, route cheap tasks to the budget pick. Benchmark and Council results feed the same intuition the Command Center tracks over time.

Where to go next

• The Model Council — rounds, formats, transcripts.
• The Benchmark suite — scoring, saved suites, history.
• OpenRouter models — widening the contender pool.