A checklist for Chinese-language model costs

Avoid a price leaderboard

A model can have different channels, versions, caching rules and batch rates. Quotes in different currencies or per-token versus per-call units are not directly comparable. Task quality and actual prompt length also affect cost.

Preserve sources for each comparison

Record the provider, full model name, channel, official pricing URL and check date. Identify units, currency, caching, minimum charges, discount conditions and failed-call billing. This article does not reproduce unverified current prices or claim universal Chinese-task coverage.

Evaluate the same tasks

Use real, redacted inputs with explicit success criteria and retry rules. Preserve returned usage and distinguish reported amounts from estimates. Unknown costs are not zero; small samples do not establish long-term cost.

AntDock can record tokens, latency, status and amount sources for instrumented calls. It cannot observe uninstrumented calls or replace invoices. Verify one call with the AI SDK reference before expanding coverage.