Among the 12 most-frequently-routed models, the top five are free. Anthropic's Claude Sonnet 4.6 — typically positioned as the default frontier reasoning model — handles 2.8 percent of all sampled calls. The two largest paid models by share are xai/grok-code-fast-1 ($0.20/$1.50 per million) and moonshot/kimi-k2.5 ($0.60/$3.00 per million).
Show more
