- Monthly calculator
- V4 Flash + Pro
- Rates checked 18 Aug 2026
Monthly calculatorDeepSeek PricingTurn token rates into a monthly bill
DeepSeek pricing depends on model, input cache hits, output tokens and time of day. V4 Flash starts at $0.007 per million cached input tokens off-peak; V4 Pro starts at $0.022. Use the calculator below for a realistic monthly estimate.
Independent calculator · Official rates checked 18 August 2026 · Prices can change.
DeepSeek introduced the current V4 peak/off-peak table on 16 August 2026. Articles using the earlier flat prices do not calculate today's V4 bill correctly.
Estimate covers DeepSeek API tokens only. It excludes taxes, your infrastructure and third-party agent or tool charges.
DeepSeek V4 pricing at a glance
- Flash cached input
- $0.007 / 1M off-peak
- Flash uncached input
- $0.22 / 1M off-peak
- Flash output
- $0.66 / 1M off-peak
- Pro cached input
- $0.022 / 1M off-peak
- Pro uncached input
- $0.66 / 1M off-peak
- Pro output
- $1.98 / 1M off-peak
- Peak premium
- 2× every token category
- Context window
- 1M tokens
Interactive calculator
DeepSeek pricing calculator for monthly API usage
Enter average tokens per request and expected monthly requests. Cache-hit and peak-traffic percentages let the result reflect how the API is actually billed.
V4 Flash / month
$112.19$0.0112 per requestV4 Pro / month
$336.88$0.0337 per requestFlash savings vs Pro
$224.69for this monthly workloadFlash peak premium
$22.44extra cost caused by peak trafficEstimate covers DeepSeek API tokens only. It excludes taxes, your infrastructure and third-party agent or tool charges.
Current API table
DeepSeek pricing per 1 million tokens
These are the V4 Flash and V4 Pro rates in DeepSeek's official documentation on 18 August 2026. Peak pricing is double the off-peak rate in every token category.
| Model and period | Cached input | Uncached input | Output |
|---|---|---|---|
| V4 Flash · Off-peak | $0.007 | $0.22 | $0.66 |
| V4 Flash · Peak | $0.014 | $0.44 | $1.32 |
| V4 Pro · Off-peak | $0.022 | $0.66 | $1.98 |
| V4 Pro · Peak | $0.044 | $1.32 | $3.96 |
V4 Flash
2,500 concurrent requestsOfficial concurrency limit shown for the Flash route.
V4 Pro
500 concurrent requestsOfficial concurrency limit shown for the Pro route.
Context
1M tokensBoth routes list a one-million-token context window.
Time-of-day pricing
When DeepSeek peak pricing applies
Peak windows are 01:00–04:00 UTC and 06:00–10:00 UTC. Requests outside those seven hours use the off-peak table.
Converted automatically
Your local peak windows are loading…Actionable savings
How to lower DeepSeek pricing without cutting output quality
The rate table exposes two high-leverage levers: schedule flexible work off-peak and improve cache reuse on repeated context.
Move batch work outside peak windows
Every token category costs half as much off-peak. Queue evaluations, indexing and long reports for the lower-rate hours.
Keep repeated prefixes stable
System prompts, tool descriptions and long shared context are more likely to benefit from caching when their prefix does not change.
Route routine work to Flash
Use V4 Pro only when your own task tests show a quality gain worth roughly three times the off-peak uncached-input and output rates.
Measure cache hit rate
A long prompt is not automatically cheap. Read the provider usage breakdown, then enter the observed hit rate in the calculator.
FAQ
DeepSeek pricing questions
DeepSeek pricing questions from the reference page.
- API tokens only
- Peak is 2×
- Verify current rates
There is no single monthly API fee in the official table. Monthly DeepSeek pricing is requests multiplied by cached input, uncached input and output charges, weighted by peak-hour traffic. Use the calculator above.
Sources
Official pricing source and limits
Independent calculator · Official rates checked 18 August 2026 · Prices can change.
- DeepSeek official pricing documentation
V4 Flash and Pro cached input, uncached input, output, peak windows, context, and concurrency.
Next step
Continue with DeepSeek
Practical pages for choosing and using agent infrastructure.