Model comparisons

Every comparison below renders from the same live catalog as the /models page, so the per-1M-token rates, context windows, and capabilities can't go stale. All of these models are callable through one OpenAI-compatible endpoint, which makes an A/B test a one-string change.