How is the monthly bill calculated?
A monthly bill is two multiplications. Per-call = input tokens × input price ＋ output tokens × output price (priced per million tokens); then × calls per day × 30 days for the monthly bill. This tool does it for every model and lines them up, cheapest first.
Why split input and output? Almost every vendor prices output higher than input, often 2–4×. So the same token count costs very differently for “long input, short output” (summarizing) versus “short input, long output” (writing). Cache hit rate covers repeated prefixes — a fixed system prompt or knowledge base — where the cached share bills at a cheaper cache price.
A support bot: ~800 input and ~200 output tokens per call, 30,000 calls a day. Say a model charges $1 input / $4 output per million tokens —
per call = (800×1 ＋ 200×4) ÷ 1M = $0.0016; a month ≈ 0.0016 × 30k × 30 = $1,440.
Switch to a model with cheaper output and that monthly bill often halves — exactly what the ranking above is for.
Chinese models (Doubao / Qwen / ERNIE / Kimi, etc.) use each vendor’s official CNY price; global models convert at the live rate, toggled ￥ / $. For each model’s context window and capabilities, see the LLM price comparison.