One page per LLM in the benchmark — its performance across every task, where it’s the cheapest qualifier at the default quality bar, and which categories it does and doesn’t qualify on.