Each model appears once, with every available group on the card. The highlighted row is the cheapest input-price group. Except per-image drawing models, prices are CNY per million tokens. The cache column shows the read price; a write price is listed only for models whose provider charges a separate cache-write fee. Model IDs must be copied exactly, including date suffixes such as claude-haiku-4-5-20251001。Setup steps are in theSetup guide. Discount = our input price ÷ the provider's official input price, using the standard list rate (excluding batch, long-context, and peak surcharges); image models use the official 1024×1024 high-quality output example.

GPT 系列长上下文与官方同步：单次请求Input（含Cache命中)over 272K token 时，整笔按官方规则计费——Input与Cache 2 倍、Output 1.5 倍。刚好 272K 不加价。

Models with a Peak badge: weekdays excluding Chinese statutory holidays, 09:00–12:00 and 14:00–18:00 (Beijing time), billed at 2×. Input, output, and cache prices all use this multiplier, matching upstream peak windows.