Contents
At a glance
- Model ID
- deepseek-v4-pro
- Context length
- 1M tokens
- Max output
- 384K tokens
- Input, cache hit
- $0.022 off-peak to $0.044 peak per 1M tokens (checked August 2026)
- Input, cache miss
- $0.66 off-peak to $1.32 peak per 1M tokens
- Output
- $1.98 off-peak to $3.96 peak per 1M tokens
- Peak hours
- 01:00-04:00 and 06:00-10:00 UTC; all other times off-peak at half the rate
- Cache discount
- Roughly 30x between cache hit and cache miss on input
What DeepSeek V4 Pro is
deepseek-v4-pro is the larger of the two model IDs on DeepSeek's API pricing page. As published in August 2026 it carries a 1M-token context length and a maximum output of 384K tokens, with all prices quoted in US dollars.
Its sibling deepseek-v4-flash shares the same 1M context and 384K output ceiling at a third of the price.
Two discounts, not one
DeepSeek's pricing has two mechanics stacked on each other, and both are unusual enough to be worth learning.
Cache hit versus cache miss. Input is priced separately depending on whether the prefix was already cached. For deepseek-v4-pro the published cache-hit rate is $0.022 to $0.044 per million tokens against a cache-miss rate of $0.66 to $1.32 per million. That is a 30x difference, far steeper than the 10x that OpenAI's caching offers, which makes prefix stability the single highest-value optimisation on this model.
Peak versus off-peak. DeepSeek's documentation states that off-peak rates are half the peak rates, with peak hours running 01:00 to 04:00 and 06:00 to 10:00 UTC and all other times treated as off-peak. That is the source of every range above: the low end of each range is the off-peak rate and the high end is peak.
Output is $1.98 to $3.96 per million tokens on the same peak and off-peak basis.
The scheduling consequence
This is a rare case where when you run a job changes what it costs. Batch work - reprocessing an archive, running an eval suite, generating summaries overnight - is literally half price if it runs outside those two UTC windows. Interactive traffic obviously cannot be moved, but most automation backlogs can.
Worth noting the windows are UTC, so a European business's working day overlaps peak considerably while a US one barely touches it. Check your own timezone against 01:00-04:00 and 06:00-10:00 UTC before assuming either way.
What you should know
Off-peak cache-hit input at $0.022 per million is roughly a tenth of the cheapest OpenAI or Google list rate, which puts DeepSeek in a different pricing class for high-repetition workloads. Whether that translates into a lower total cost depends on your prefix hit rate, which depends on whether your system prompt and tool definitions are genuinely stable between calls.
The non-price considerations are real and belong in the decision: where inference runs, what the retention policy is, and whether that is compatible with your obligations. See data residency, zero data retention and DPA before putting customer data through any API on price alone.
DeepSeek V4 Pro FAQ
How much does deepseek-v4-pro cost?
Input on a cache miss is $0.66 to $1.32 per million tokens and output is $1.98 to $3.96, with the low end of each range being the off-peak rate. Cache-hit input is $0.022 to $0.044 per million. Figures as published in August 2026.
What are DeepSeek's peak hours?
DeepSeek documents peak hours as 01:00 to 04:00 and 06:00 to 10:00 UTC, with all other times treated as off-peak at half the peak rate. Batch work scheduled outside those two windows costs literally half as much.
How big is the DeepSeek cache discount?
Around 30x on input - $0.022 to $0.044 per million on a cache hit against $0.66 to $1.32 on a miss. That is far steeper than the 10x typical elsewhere, which makes keeping the request prefix byte-stable the highest-value optimisation on this model.
What is the context window of DeepSeek V4 Pro?
1M tokens, with a maximum output of 384K tokens, as published on DeepSeek's pricing documentation in August 2026. DeepSeek V4 Flash carries the same figures at a lower price.
Is DeepSeek cheaper than OpenAI?
On list price for high-repetition workloads, substantially - off-peak cache-hit input at $0.022 per million is roughly a tenth of the cheapest OpenAI or Google list rate. Whether that holds for you depends on your actual cache hit rate, and price is not the only factor: check data residency, retention and your DPA obligations first.
Can I schedule work to get the off-peak rate?
Yes, and it is one of the few genuine scheduling arbitrages in this market. Archive reprocessing, eval suites and overnight summarisation can all run outside 01:00-04:00 and 06:00-10:00 UTC at half price. Interactive traffic cannot be moved, so split the two.
Primary source
Checked August 2026. Anything priced or rate-limited moves, so verify against the source before you budget.
Want this built rather than explained?
Knowing what deepseek v4 pro means is one thing; having it running in your business is another. We scope one workflow at a time, quote a fixed price in writing after a free consult, build it, and hand it over with documentation you own.
