Estimate cost from usage.
Paste reported token counts and enter your provider's applicable prices. No key or model request.
Your estimate
Paste usage, enter the applicable prices, then estimate.
How the estimate works
Input is divided into ordinary tokens, cached reads and cache writes. Reported 30-minute and 12-hour writes use separate prices; their aggregate is not added again. Reasoning tokens already included in output are not counted twice.
Choose the actual usage format. DeepSeek's cache hit/miss fields differ from OpenAI's detail objects. A missing cache count stays unknown until you explicitly apply the zero assumption. An unsplit write count needs your chosen lifetime and rate.
Enter prices in one currency, per million tokens, for the actual model, context range and service tier. Currency changes clear prices. This worksheet does not fetch prices or verify an invoice; reported audio/image tokens and unmapped token counters need a different pricing calculation.
Sources checked 25 October 2026: Chat Completions usage, cache accounting and write lifetimes, and DeepSeek cache fields.
Need the original counts? Capture a full response or inspect saved JSON/SSE. Compare access conditions and dated host listings on API access comparison.