Blended $/1M tokens. input×0.75 + output×0.25. A single, stated mix so models compare apples-to-apples instead of by a misleading input-only price.

Price-performance frontier. A model is on the frontier if no cheaper model is at least as good (Pareto). Quality is an explicit, disclosed index — never hidden.

GPU normalization. We show $/hr and $/hr per 100GB VRAM so a big card and a small card compare fairly.

Data & honesty. Prices come from public sources, refresh on a timer, and fall back to the last-good cache so the site never shows stale-empty. Affiliate links are marked rel="sponsored nofollow". This is information, not advice; verify on the provider before you spend.