Plan limits

What does the DeepSeek V4 Pro API cost, and what are its rate limits?

DeepSeek V4 Pro is $0.66 per million input tokens and $1.98 per million output off-peak, double that at peak. One account gets 500 concurrent requests, and DeepSeek says the model keeps running after September 14, 2026.

Checked against the lab's own pages on September 15, 2026. Limits change often, so the date matters.

DeepSeek's own pricing page puts deepseek-v4-pro at $0.66 per million input tokens on a cache miss and $1.98 per million output tokens, off-peak. At peak it's double: $1.32 and $3.96. A cache hit is $0.022 off-peak, and $0.044 at peak.

Model version

DeepSeek-V4-Pro-0813

Off-peak

$0.66 input on a cache miss, $1.98 output

Peak

$1.32 input on a cache miss, $3.96 output

Cache hit

$0.022 off-peak, $0.044 peak

Peak hours

01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday

Concurrency limit

500 per account, HTTP 429 above it

When peak actually is

Peak is narrower than you'd guess. "Peak hours are 01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak)." That's seven hours a day, five days a week. Everything else, weekends included, bills at the lower rate.

DeepSeek writes it the other way round on the page: "Off-peak rates are half of the peak rates." Same arithmetic either way, and it's worth knowing which side of it a nightly batch job lands on.

Concurrency, and what happens over it

The concurrency limit is 500 for deepseek-v4-pro, against 2500 for deepseek-flash. "A request counts as one concurrent connection from the time it is sent until the model response is complete", and the count is per account rather than per API key. Go past it and you get an HTTP 429 (the docs don't mention a queue).

You can ask for more. DeepSeek takes capacity expansion requests, matches the concurrency to your business needs, and says there's "no additional cost for capacity expansion". Granted, 500 in flight at once is already a lot unless you're fanning out agents.

The model is staying

The Flash line is the one that moved. deepseek-flash is the name to use now; the legacy deepseek-v4-flash and deepseek-v4-flash-vision-exp names still get accepted, but those models are retired and their requests are served by DeepSeek-V4.1-Flash and billed at the Flash price.

What V4 Pro gives up

Vision works on deepseek-flash and is "Not supported" on deepseek-v4-pro, which is the one real feature gap between them. Both carry a 1M-token context and a maximum output of 384K, both do JSON output and tool calls, and both take requests in either the OpenAI or the Anthropic format.

Charges come off your topped-up balance or your granted balance, granted balance first when you have both. DeepSeek says prices may vary and asks you to check the page.