DeepSeek API for free: is there a free tier and what is the minimum cost

The query "deepseek api free" usually means one of two things: a person wants to try it without paying, or hopes never to pay at all. The honest answer is short: through Zukko there is no free access, the API works on a pay-per-token basis. But DeepSeek tokens cost so little that the entry threshold is measured in cents. Below we count it in numbers.

Is the DeepSeek API free

Model weights and a chat site with the models are one thing. API access, meaning calls to a model from your code, is another: servers stand behind it, and someone pays for them. That is why an API almost always has a per-token tariff. The terms of the official DeepSeek platform (including any trial credits) change and we do not know them, so check the current information on the service's own site.

How it works with us:

• one dollar balance, you pay only for tokens spent, no subscription;
• the minimum top-up is $1;
• there is no welcome credit for signing up;
• there is a single one-off credit: $0.50 for subscribing to the @zukkopro↗ channel, once per Telegram account (link Telegram in the dashboard, subscribe, press "Check").

A separate warning about "free DeepSeek keys" posted in chats and forums. Someone else's key can stop working at any moment, and your requests and data pass through an unknown owner. There is one working path: your own key and your own balance.

DeepSeek prices in our table

Prices as of 18 October 2026, dollars per 1M tokens:

For comparison, the same figures for other cheap models: glm-5.3-flash 0.04 / 0.04, qwen3.8-flash 0.038 / 0.075. The "strength index" is the Intelligence Index from Artificial Analysis; we give no other quality figures, so check whether a model fits your task yourself.

The current list and prices are always returned by GET https://api.zukko.pro/v1/models.

How many requests $1 buys

Take a typical small request: 2000 input tokens (a question and a little context) and 500 output tokens. The formula:

request cost = input × input price / 1,000,000 + output × output price / 1,000,000

DeepSeek V4.1 Flash: 2000 × 0.013 / 1,000,000 = 0.000026 500 × 0.038 / 1,000,000 = 0.000019 Total $0.000045 per request. Divide a dollar by that: 1 / 0.000045 ≈ 22,200 requests.

DeepSeek V4 Pro: 2000 × 0.075 / 1,000,000 = 0.00015 500 × 0.225 / 1,000,000 = 0.0001125 Total $0.0002625. 1 / 0.0002625 ≈ 3,800 requests.

Summary for the same request:

For glm-5.3-flash: 2000 × 0.04 + 500 × 0.04 = 100,000, divided by 1,000,000 gives $0.0001. For claude-sonnet-5-5: 2000 × 0.4 + 500 × 2 = 1800, divided by 1,000,000 gives $0.0018.

What changes with large contexts

The calculation above is for short requests. If you put a 100,000-token document into the request, input starts to dominate:

100,000 × 0.013 / 1,000,000 = $0.0013 for the input of a single request alone.

For DeepSeek the cache read price equals the input price (0.013 and 0.013), so the cache makes nothing cheaper here. The gateway sees the cache if the provider reports it in the response, and cache writes are not billed separately. If savings on a repeated long prefix matter to you, compare this scheme against other models; details are in the docs on limits and cache↗.

How to top up without overpaying

• SBP (Russian instant payments). A payment in rubles at the market rate at the moment of payment; the payment page adds an 8.5% fee on top. The balance receives the dollar amount you requested. Topping up $1 costs the equivalent of 1 × 1.085 = $1.085.
• Cryptocurrency (USDT and others).
• MasterCard cards. A fee of about 13%, amounts from $2 to $594.

For an experiment the minimum top-up is enough: it covers thousands of V4.1 Flash requests. A step-by-step SBP guide is in the post on paying via SBP↗.

The first request

from openai import OpenAI

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

r = client.chat.completions.create(
    model="deepseek-v4.1-flash",
    messages=[{"role": "user", "content": "Explain prompt caching in two sentences"}],
)
print(r.choices[0].message.content)
print(r.usage)  # how many tokens were actually charged

The usage field shows the exact token count, and the "Usage" page in the dashboard shows the cost. So that a stray loop does not spend more than planned, set a daily dollar limit on the key (it is configured in the same place where the key is created). Create a key in the dashboard↗. More on accessing the model from Russia is in the post on DeepSeek in Russia↗, and the model list is described in the models reference↗.

FAQ

Does the DeepSeek API have a free tier?

Through Zukko there is no free tier: you pay per token from a shared balance. For terms on the official DeepSeek platform, check its site, they change.

How much does it cost to try?

The minimum top-up is $1. By the calculation above that is about 22,000 short requests to deepseek-v4.1-flash, but your requests will have their own size, so check the usage field.

Is the DeepSeek API paid or not if I make one or two requests a day?

Paid, but the sums are tiny: at $0.000045 per request, 60 requests a month cost about $0.0027.

Will the cheapest model suit my task?

The strength index of deepseek-v4.1-flash is 39 and of deepseek-v4-pro 36 (unlike Flash, it accepts images). A number says nothing about your task: run 20 typical examples on two models and compare the results.