From prompt to request cost
Estimated request cost equals input tokens multiplied by $0.435 per million, plus expected output multiplied by $0.87 per million. Cached input is separate only when the provider publishes a rate.
DeepSeek token calculator
Plan DeepSeek requests before they run.
Estimate prompt size, output cost, and context pressure for DeepSeek routes without uploading your text.
DeepSeek V4 Pro1,048,576token context window
$0.435 input / 1M $0.003625 cached / 1M $0.87 output / 1M
Official · verified 2026-08-16View provider pricing ↗Automatic disk caching makes cache hits roughly 1/120th of the fresh input price.What this page helps you decide
- Compare input and output cost on the same prompt.
- Reserve output before checking context fit.
- Verify the final bill against DeepSeek usage data.
Prices are planning references, not a billing guarantee. Check the provider’s current pricing before launch.
Estimate a DeepSeek V4 Pro request
1,048,576 context
0input tokens · local estimate
$0.00estimated input cost
$0.0004reserved output cost
$0.0004estimated request total
<0.1% of context reserved. Close approximation · o200k proxy — counts are exact for OpenAI models and a close approximation for DeepSeek; provider formatting, tools, and files can change the real count.
Count the assembled request
System instructions, conversation history, retrieved documents, tool definitions, files, and response length all matter. Use the visible prompt for drafting, then validate the final payload with DeepSeek.
Before you rely on the number
DeepSeek V4 Pro calculator questions
Is this DeepSeek V4 Pro token count exact?
No. It is a private plain-text planning estimate. DeepSeek tokenization, chat formatting, tools, files, and images can change the real count.
Does the prompt leave my browser?
No. The calculator runs locally in your browser. Only the page itself and published registry data are loaded from PromptCostLab.
How should I verify production cost?
Check the linked DeepSeek pricing page, then compare the estimate with token usage returned by the real API request.