Highlights shows the true top 10 models by intelligence - no per-company limit, so if one company genuinely has several of the top models right now, several of its models are shown. Ranked three ways - Intelligence, Price, and Speed - so you can see how the same lineup trades off against each other. Each model keeps one color across all three charts. Value (0-100) below that shows how much intelligence you get per dollar, scaled so the best value model shown scores 100 - tables sort by Value by default; click "Intelligence" or "Name" above one to sort by those instead. The three calculated columns are, in full: Cost / task = price per 1M tokens × tokens the model actually uses ÷ 1,000,000 Value = intelligence score ÷ cost per task Price = (3 × input + output) ÷ 4, per 1M tokens The Best Value Pick is the sharpest bend in the price-to-intelligence curve across every model in the category. The table also includes every other model on that same efficiency frontier (tagged Frontier) even if it didn't score high enough to crack the top 10 - a cheap, well-balanced model would otherwise be invisible no matter how the table is sorted. Use the Capability buttons to control how far down that list goes - All (default) filters nothing; Flagship keeps models within 10% of the best score, Strong within 20%, Broad within 50%. The Best Value Pick always stays visible regardless of which tier is selected. Click Update on a category page (or Update All on the home page) to refetch live data any time.