DeepSeek API cost per request, October 2026: a chat turn and a long document

DeepSeek bills API calls per token, at three rates per million tokens: input, cached input and output. The Flash models cost cents per million tokens and Pro costs under a dollar in, so the model you pick moves the bill more than anything else. A request costs its tokens times the rates of the model it named; there is no seat or plan price.

7 DeepSeek models. Each row shows our price and DeepSeek's list price.

The lowest price here is DeepSeek V4 Flash 0423: $0.0279 in and $0.0558 out.

A chat turn of 2,000 input and 500 output tokens, and a long-document request of 50,000 input and 2,000 output. Both sizes are assumptions; nothing is counted at the cached rate.

DeepSeek V4 Pro 0423: $0.792 in and $1.92 out here, against $1.32 and $3.96 at list, 40% under on input and 51% on output.

DeepSeek V4.1 Flash: $0.06 in and $0.18 out here, against $0.30 and $1.20 at list, 80% under on input and 85% on output.

DeepSeek V4 Flash 0423: $0.0279 in and $0.0558 out here, against $0.30 and $1.20 at list, 90% under on input and 95% on output.

Two monthly workloads on the same models: 1,000 chat turns a day and 100 long documents a day, each for 30 days. Every request is the size above, so a month of chat is 60 million input and 15 million output tokens, and a month of documents is 150 million input and 6 million output. The volumes are assumptions; nothing is counted at the cached rate.

DeepSeek V4 Pro 0423: $206.64 a month here for both workloads, against $360.36 at list, $153.72 less.

DeepSeek V4.1 Flash: $16.38 a month here for both workloads, against $88.20 at list, $71.82 less.

DeepSeek V4 Flash 0423: $7.03 a month here for both workloads, against $88.20 at list, $81.17 less.

Cost scales with volume, so 10,000 chat turns a day is ten times the chat column.

When a request repeats a prefix the API has seen, such as a long system prompt or a document, the response reports those tokens as cached and they bill at the cached rate in the first table. A model with no cached rate bills them at its input rate.

Point the SDK at our base URL with your key. Every DeepSeek id in the table works with the same two values.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.uplinkswitch.com/v1",
    api_key="sk-uplink-live-...",
)

completion = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[{"role": "user", "content": "Summarize this in one line."}],
)
print(completion.choices[0].message.content)

More in the SDK guide.

Is there a subscription? No. Credit is prepaid from $10, and each top-up locks your prices for 30 days.

Which DeepSeek model should I use? Flash for chat, summaries and agent subtasks, where the price per million is low enough to stop counting. Pro for code and reasoning where Flash gets it wrong. Run the same prompt on both and compare.

Where is the full price list? On the DeepSeek API pricing page, and every other model on the Models page. To see what your own usage would cost, re-price an export on the savings page.