DeepSeek V4 Pro costs $0.66 per million input tokens off-peak and $1.32 at peak on the official DeepSeek API, with output at $1.98 and $3.96. V4.1 Flash, the lower-cost model in the family, costs $0.15 and $0.30 input and $0.60 and $1.20 output. Both bill by time of day, with off-peak at half the peak rate.
The official API is only one of the places V4 Pro is sold. Twenty providers list the same model from $0.25 to $1.65 per million input tokens, and which one is cheapest depends on the hour you run and how much of your input is cached. This page covers the official rates, the full provider table, what a real workload pays, and how to choose.
One correction to the record. On September 10, 2026, DeepSeek said V4 Pro requests would be routed to V4.1 Flash from September 14. It reversed that decision before the date, and V4 Pro remains on the official API at the same billing.
For the V4 Pro deep dive, see DeepSeek V4 Pro API pricing. For the newer model, see DeepSeek V4.1 Flash API pricing. For every DeepSeek model and host in one place, see DeepSeek API pricing.
What DeepSeek V4 is, briefly
DeepSeek V4 Pro was released on April 24, 2026 as the flagship successor to the V3 line. It is an open-weight mixture-of-experts model with 1.6 trillion total parameters and 49 billion active, a 1M token context window, tool calling, and prompt caching. The general availability build is V4 Pro 0813.
It launched alongside DeepSeek V4 Flash, which was retired on September 10, 2026 and replaced by V4.1 Flash. The legacy V4 Flash model names are still accepted on the official API, but requests are served by V4.1 Flash and billed at the Flash price.
DeepSeek V4 official API pricing
Rates from DeepSeek's pricing page, USD per 1M tokens:
| Token type | V4 Pro off-peak | V4 Pro peak | V4.1 Flash off-peak | V4.1 Flash peak |
|---|---|---|---|---|
| Input (cache miss) | $0.66 | $1.32 | $0.15 | $0.30 |
| Input (cache hit) | $0.022 | $0.044 | $0.003 | $0.006 |
| Output | $1.98 | $3.96 | $0.60 | $1.20 |
Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday. All other hours, including weekends and Chinese public holidays in full, bill at off-peak rates, which are half of peak. The hours in your time zone are on DeepSeek peak and off-peak hours.
These rates took effect on August 16, 2026. Before that, V4 Pro ran at a flat $0.435 input and $0.87 output.
DeepSeek V4 Pro pricing by provider
Twenty providers list the V4 Pro 0813 build across 22 listings on OpenRouter as of October 13, 2026, sorted by input price. USD per 1M tokens.
| Provider | Input | Output | Cached input |
|---|---|---|---|
| Wafer | $0.25 | $5.00 | $0.20 |
| Ionstream | $0.37 | $2.93 | $0.088 |
| Sail Research (two listings) | $0.48 | $1.98 | $0.10 |
| DeepSeek (official, off-peak) | $0.66 | $1.98 | $0.022 |
| StreamLake (50% off promotion) | $0.66 | $1.98 | $0.022 |
| NovitaAI (25% off promotion) | $0.99 | $2.97 | $0.033 |
| GMICloud (20% off promotion) | $1.056 | $3.168 | $0.0352 |
| Alibaba Cloud International (second listing) | $1.122 | $3.366 | $0.1122 |
| Reka AI | $1.20 | $3.20 | $0.06 |
| DeepInfra | $1.30 | $2.60 | $0.10 |
| CoreWeave | $1.31 | $3.96 | $0.044 |
| DeepSeek (official, peak) | $1.32 | $3.96 | $0.044 |
| Alibaba Cloud International | $1.32 | $3.96 | $0.132 |
| AtlasCloud | $1.32 | $3.96 | $0.132 |
| Baidu Qianfan | $1.32 | $3.96 | $0.042 |
| Parasail | $1.32 | $3.96 | $0.044 |
| Together | $1.32 | $3.96 | $0.13 |
| DigitalOcean | $1.32 | $3.96 | $0.044 |
| SiliconFlow | $1.32 | $3.96 | $0.044 |
| Cloudflare | $1.32 | $3.96 | $0.044 |
| Phala | $1.45 | $4.36 | $0.15 |
| Venice | $1.65 | $4.95 | $0.165 |
Three things to read from this table.
The official peak rate is the reference price. Eight third-party hosts list at exactly $1.32 input and $3.96 output, the same as DeepSeek at peak, and none of them offer an off-peak discount. A buyer on one of those hosts pays the peak rate around the clock.
A few hosts now undercut the official off-peak input rate. Wafer, Ionstream, and Sail Research all list below $0.66. No host is below the official off-peak output rate of $1.98, and Wafer's $5.00 output is the highest on the table.
Three listings are promotions. StreamLake, NovitaAI, and GMICloud carry marked discounts of 50, 25, and 20 percent. The list rate behind all three is the official peak rate. Treat the discounted price as temporary.
OpenRouter also lists the April preview build, V4 Pro 0423, as a separate model with its own hosts and prices. Confirm which build a host serves before comparing its rate to this table.
Where the official API still wins
When this page was first written, the official DeepSeek API was the cheapest of all 16 hosts. That is no longer true on uncached input, but it is still true where most production traffic sits: cached input.
DeepSeek charges $0.022 per million cached input tokens off-peak. The cheapest third-party hosts charge $0.088 to $0.20, four to nine times more. For an agent workload where most of the prompt repeats, that gap outweighs a lower headline input price.
The lesson for buyers is that official versus third-party has no fixed answer. It changes by model, by hour, and by cache hit rate, which is why rate-card assumptions carried from one model to another keep being wrong. The pattern across the whole market is in why token prices differ.
What a real workload pays
A production assistant doing 10M input and 1M output tokens per day on V4 Pro, with no caching:
| Where it runs | Daily | Annual |
|---|---|---|
| Ionstream ($0.37 / $2.93) | $6.63 | $2,420 |
| Sail Research ($0.48 / $1.98) | $6.78 | $2,475 |
| DeepSeek official, entirely off-peak | $8.58 | $3,132 |
| DeepSeek official, spread evenly around the clock | $10.37 | $3,784 |
| DeepSeek official, entirely at peak | $17.16 | $6,263 |
| Any host at the peak rate ($1.32 / $3.96) | $17.16 | $6,263 |
Now add a 60 percent cache hit rate on input:
| Where it runs | Daily with 60% cached input |
|---|---|
| Sail Research ($0.10 cached) | $4.50 |
| DeepSeek official, off-peak ($0.022 cached) | $4.75 |
| Ionstream ($0.088 cached) | $4.94 |
| DeepSeek official, peak ($0.044 cached) | $9.50 |
At 60 percent cached, the three cheapest options are within fifty cents a day of each other. Push the cache hit rate to 90 percent and the official API off-peak pulls ahead, at about $2.84 a day against about $3.36 at Sail Research.
If the workload does not need V4 Pro, V4.1 Flash runs the same uncached 10M and 1M day for $2.10 off-peak and $4.20 at peak on the official API.
DeepSeek V4 Pro vs V4.1 Flash
| V4 Pro | V4.1 Flash | |
|---|---|---|
| API model name | deepseek-v4-pro | deepseek-flash |
| Input, off-peak and peak | $0.66 and $1.32 | $0.15 and $0.30 |
| Output, off-peak and peak | $1.98 and $3.96 | $0.60 and $1.20 |
| Cached input, off-peak and peak | $0.022 and $0.044 | $0.003 and $0.006 |
| Context window | 1M tokens | 1M tokens |
| Max output | 384K tokens | 384K tokens |
| Vision | Not supported | Supported |
| Concurrency limit | 500 | 2,500 |
V4.1 Flash costs about a quarter of V4 Pro on input and under a third on output, adds vision, and has five times the concurrency limit. DeepSeek says it also beats V4 Pro on performance and speed. Test it on your own task before assuming V4 Pro is required.
For how DeepSeek compares with other model families, see the LLM API pricing comparison.
How to choose a V4 provider
- Uncached traffic: Ionstream or Sail Research, which undercut the official API at every hour.
- Cache-heavy traffic: the official API off-peak. Its cached input rate is the lowest on the table by a wide margin.
- Traffic you can schedule: move it off-peak on the official API and the rate halves with no other change.
- No reasoning required: V4.1 Flash at $0.15 and $0.60 off-peak.
- Compliance, region, or platform needs: price the premium explicitly. Most enterprise-oriented hosts charge the full peak rate at all hours.
- Budget certainty: DeepSeek has repriced by announcement once this year and changed and reversed a routing plan within days. For predictable volume, see how to lock in token prices.
- Recheck monthly. In six weeks this market went from 16 hosts with the official API cheapest to 20 providers with three below it.
To track token prices across models and hosts, see the Mercatus Token Index.
Frequently asked questions
How much does DeepSeek V4 cost per million tokens?
As of October 13, 2026: V4 Pro costs $0.66 input and $1.98 output off-peak, doubling to $1.32 and $3.96 at peak hours (01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday). V4.1 Flash costs $0.15 and $0.60 off-peak, $0.30 and $1.20 at peak. Cached input runs $0.022 to $0.044 for Pro.
Who is the cheapest DeepSeek V4 provider?
For uncached workloads, Ionstream ($0.37 input, 0.48 input, $1.98 output). For cache-heavy workloads, the official DeepSeek API off-peak, because its cached input rate of $0.022 is far below any third-party host.
Is DeepSeek V4 Pro still available on the official API?
Yes. DeepSeek announced on September 10, 2026 that V4 Pro requests would be routed to V4.1 Flash from September 14, then reversed the decision. V4 Pro remains available as deepseek-v4-pro with unchanged billing.
What is the difference between V4 Pro and V4.1 Flash?
V4 Pro is the larger flagship at $0.66 and $1.98 off-peak. V4.1 Flash is the newer, lower-cost model at $0.15 and $0.60 off-peak, with vision support and a higher concurrency limit. Both have a 1M token context window. V4 Flash, the original budget model, was retired on September 10, 2026.
Why do many third-party hosts charge the official peak rate?
Eight hosts list V4 Pro at exactly $1.32 input and $3.96 output, matching DeepSeek's peak price. Third-party hosts carry their own GPU costs and margin, and most do not offer time-of-day discounts, so they anchor to the official peak rate and compete on enterprise terms, regions, and platform features instead.
How wide is the price spread across V4 Pro providers?
Input runs from $0.25 to $1.65 per million tokens, about 6.6 times. Output runs from $1.98 to $5.00, about 2.5 times.
Methodology
Official rates, peak hours, model names, concurrency limits, and feature support from DeepSeek's API pricing page, read October 13, 2026. Third-party listings from OpenRouter's DeepSeek V4 Pro 0813 page, standard routing, read October 13, 2026; promotional listings are shown at the discounted rate with the discount noted. Release date and parameter counts from OpenRouter's model listing. Pre-increase rates and the September routing announcement and reversal are from DeepSeek's documentation and change log as recorded on earlier versions of this page. Workload examples use stated token volumes and cache assumptions at 365 days; the around the clock figure assumes 35 of 168 weekly hours at peak. Recompute with your own mix. Last verified: 2026-10-13.