Batch summaries
100k input tokens and 50k output tokens for a content batch.
DeepSeek calculator
DeepSeek Token Calculator
Estimate uncached DeepSeek API spend with the current V4 Flash and V4 Pro rates from the shared pricing store.
Prices verified: via OpenRouter
Interactive tool
Convert tokens to cost
Use presets, share the exact inputs, and scan the live breakdown.
Estimated total cost$0.028
Cost per 1k tokens$0.0002
Examples
Short support reply
2,000 prompt tokens and a compact 300-token answer.
DeepSeek V4 Flash: $0.14 input / $0.28 output per 1M tokens.
How this DeepSeek Token Calculator works
DeepSeek V4 Flash and V4 Pro price tiers
DeepSeek's current text API presents a compact two-model choice. V4 Flash is the throughput-oriented tier at $0.14 input and $0.28 output per 1M tokens; V4 Pro raises the standard rate to $0.435 input and $0.87 output per 1M tokens for harder reasoning, coding, and long-horizon work. Both endpoints support thinking and non-thinking modes, so the bill is shaped less by a separate reasoning SKU than by which tier you select and how many completion tokens the task consumes. Choose Flash for chat, routing, extraction, and frequent code assistance. Reserve Pro for requests where stronger reasoning can prevent retries or replace a multi-call workflow.
Why DeepSeek output matters even at low rates
For both listed V4 models, the standard output rate is exactly twice the cache-miss input rate. That makes response length the first lever to watch: one generated token costs the same as two fresh prompt tokens. A retrieval workflow with a large document and a short answer can remain remarkably inexpensive, while an agent that reasons verbosely, rewrites files, or produces several drafts shifts more of its spend to output. The calculator separates the two meters so a shorter completion limit can be tested without pretending the prompt shrank. Thinking tokens reported as output also belong in the output estimate when measuring real calls.
Cache hits change the DeepSeek bill
DeepSeek automatically discounts input tokens that match reusable context in its disk cache. This page deliberately uses the standard cache-miss input rate from the central store, giving a conservative baseline for a new or changing prompt. Repeated system instructions, reference material, and conversation prefixes can make the invoice lower than this estimate, but the hit rate depends on real traffic and should not be guessed in a planning calculator. Read the API usage fields for cache-hit and cache-miss tokens, then model a measured effective input cost separately. Output receives no equivalent cache discount, so long answers remain the stable cost center.
When DeepSeek is cheap—and when Pro can be expensive
V4 Flash is a strong cheap choice for high-volume text processing, especially when outputs are compact and repeated context earns cache hits. V4 Pro is the expensive option inside DeepSeek's own lineup: its input rate is more than three times Flash's, so upgrading every request can erase the value of routing. It can still be economical against premium frontier APIs when difficult work succeeds in one call. Test quality on your own data, route routine requests to Flash, and promote only the cases that benefit from Pro. Tool fees, retries, regional taxes, and any third-party gateway markup remain outside this token-only estimate.
Examples
DeepSeek V4 Pro · 1,000,000 input and 250,000 output tokens. From the central pricing store ($0.435 input and $0.87 output per 1M tokens), input costs $0.435, output costs $0.2175, and the total is $0.6525.
FAQ
Does this DeepSeek calculator include cache-hit pricing?
No. It uses the standard cache-miss input rate so the estimate remains conservative. DeepSeek reports cache-hit tokens separately, letting you replace the baseline with measured production data later.
Should I choose DeepSeek V4 Flash or V4 Pro?
Start with V4 Flash for frequent, latency-sensitive, or straightforward work. Use V4 Pro where harder reasoning or coding quality offsets its higher token rate and reduces retries.
Are thinking tokens free on DeepSeek V4?
No. Thinking and generated answer tokens contribute to billed output usage. Capture the API's actual usage fields when estimating a reasoning-heavy production workflow.