DeepSeek API price: cost of V4 Pro and V4.1 Flash worked through

The DeepSeek API price per single request is tiny, which makes it easy to underestimate or overestimate: the tariff has a string of zeros, while a project counts requests in the tens of thousands. Below we calculate the cost of three models, deepseek-v4-pro, deepseek-v4.1-flash and deepseek-v4-flash, at Zukko prices on 18 October 2026, with the formula in view and a comparison with the vendor's official prices. Prices change often; the current list is at GET https://api.zukko.pro/v1/models.

Prices per million tokens

All prices are in dollars per 1M tokens.

"Strength index" is the Intelligence Index from Artificial Analysis; we give no other quality figures. Note that deepseek-v4.1-flash has a higher index than deepseek-v4-pro and is cheaper, but it does not accept images. Whether it suits your task, check on your own data. For connection details see the article on DeepSeek V4.1 Flash via API↗.

The formula

Request cost:

input × input price + output × output price, with tokens divided by one million.

If part of the input came from the cache, the formula extends: fresh input × input price + cache read × read price + output × output price. For DeepSeek models in the table the cache read price equals the input price, so in Zukko calculations the cache does not change the total. Cache writes for these models are not billed separately. How the cache works for other families is explained in the limits and cache documentation↗.

A step-by-step example for deepseek-v4.1-flash, a request with 2,000 input tokens and 500 output tokens:

• input: 2,000 × $0.013 / 1,000,000 = $0.000026
• output: 500 × $0.038 / 1,000,000 = $0.000019
• total: $0.000045 per request, that is $0.45 per 10,000 requests.

At official prices the same request: 2,000 × $0.15 / 1M + 500 × $0.6 / 1M = $0.0003 + $0.0003 = $0.0006, and 10,000 requests come to $6.00.

Three typical scenarios at 10,000 requests

Assumptions: input and output tokens are set in each row, the cache is not counted (per the table it does not change the price). See real sizes in the usage field of a response. Russian text usually has more tokens than English for the same meaning.

Chatbot: 2,000 input, 500 output. Answering from documents (RAG): 12,000 input, 800 output. Batch classification: 1,500 input, 200 output.

Total for 10,000 requests, in dollars (Zukko / official prices):

Checking one cell, RAG on deepseek-v4-pro: input 12,000 × $0.075 / 1M = $0.0009, output 800 × $0.225 / 1M = $0.00018, total $0.00108 per request, × 10,000 = $10.80. At official prices: 12,000 × $0.66 / 1M = $0.00792, 800 × $1.98 / 1M = $0.001584, total $0.009504, × 10,000 = $95.04.

You can see that cost is almost entirely set by the input when requests are long (RAG). So the first lever is not to send extras: trim history and select only the needed document fragments. Tips are collected in how to cut token usage↗.

A script: compute cost from usage

Do not guess tokens: every response contains usage. A Python example with the official openai SDK:

from openai import OpenAI

client = OpenAI(base_url="https://api.zukko.pro/v1", api_key="fetch_your_key")

PRICES = {  # $ per 1M tokens: input, output, cache read (as of 2026-10-18)
    "deepseek-v4-pro": (0.075, 0.225, 0.075),
    "deepseek-v4.1-flash": (0.013, 0.038, 0.013),
    "deepseek-v4-flash": (0.025, 0.063, 0.025),
}

def cost(model, usage):
    p_in, p_out, p_cache = PRICES[model]
    details = getattr(usage, "prompt_tokens_details", None)
    cached = (getattr(details, "cached_tokens", 0) or 0) if details else 0

You will have to update the PRICES dictionary when prices change. For reliable reporting, check the spending history in the dashboard ("Usage") or via the API (GET /v1/usage). You can estimate input size in advance without spending anything with a count_tokens request (see the documentation↗).

Comparison with official prices: what to remember

In the table above the dollar difference is visible directly in the numbers; we derive no multipliers. Three caveats:

1. Official prices are taken as of 18 October 2026 and may change. Check the vendor's site.
2. The vendor may price the cache differently; we do not compare official cache prices here.
3. Real savings depend on your volumes. At 100 requests a day the difference is cents, at a million it is hundreds of dollars.

Limits also affect the budget: you can set a daily and monthly dollar limit on a key, and even a bug in a loop will not go further.

How to top up

The balance is kept in dollars, top-up from $1. Payment from Russia is available via SBP (the payment page adds an 8.5% commission on top: for $10 on the balance you pay $10 × 1.085 = $10.85 in roubles at the market rate at the moment of payment), by MasterCard (a commission of about 13%, amounts from $2 to $594) or with cryptocurrency. Create a key in the dashboard↗.

FAQ

How much does a million tokens of DeepSeek V4.1 Flash cost?

On 18 October 2026 in Zukko: $0.013 per 1M input tokens and $0.038 per 1M output tokens. The price and model list is current at GET /v1/models.

Is the cache cheaper on DeepSeek models?

Per the Zukko table, the cache read price for DeepSeek equals the input price, so the calculation has no cache discount. Cache writes are not billed separately. For Claude and GPT the cache works differently; see the documentation.

How does DeepSeek V4 Pro differ from V4.1 Flash in price and capabilities?

V4 Pro costs more ($0.075 / $0.225 versus $0.013 / $0.038) and accepts image input; V4.1 Flash does not accept images but is cheaper. Strength index: 36 for V4 Pro and 39 for V4.1 Flash. Which model is better for your task, check on your own data.

How do I stay within budget?

Set a daily and monthly dollar limit on the key, cap max_tokens, and regularly look at "Usage" in the dashboard. When a key limit is reached, a 429 error is returned.