Does DeepSeek have a subscription? How API billing works

No monthly plan for the API. DeepSeek's API pricing page (checked 26 October 2026) lists per-token prices and says fees are deducted from a topped-up or granted balance. There are no tiers on it, no “Pro plan”, no renewal date. Your monthly price is simply what you send in a month, and below you will work that out for a small bot.

A balance, not a contract#

Think of a prepaid SIM card rather than a phone contract. You put money in, each call takes a few fractions of a cent out, and when you stop calling, the spending stops too. Nothing renews on the 1st, and there is nothing to cancel. The pricing page adds one detail: if you have a granted balance (credit the account was given), it is used first, then your topped-up money.

The DeepSeek chat app is a separate product with its own terms; check them in the app. This guide is about the API, because that is where a bill shows up. If “token” still sounds like jargon, what is a token takes two minutes.

What a token costs at DeepSeek#

The page lists two models. deepseek-flash is served by DeepSeek-V4.1-Flash-1017, and deepseek-v4-pro by DeepSeek-V4-Pro-0813. The older names deepseek-v4-flash and deepseek-v4-flash-vision-exp still work and are billed at the Flash price. Context is 1M tokens, maximum output 384K.

Prices per 1M tokens, in US dollars, from the same page on 26 October 2026:

Two things to read out of it. First, peak is double off-peak. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, except Chinese public holidays, so the same request can cost twice as much depending on when your bot wakes up. Second, “cache hit” means DeepSeek has already seen the start of your prompt and charges far less for it; what is prompt caching shows how that works. And as in every API, output tokens cost more than input ones, which the input vs output guide explains.

DeepSeek may adjust these prices, so check the page before you plan a budget. For a deeper look at the numbers, see DeepSeek API pricing in 2026.

A month for a small bot#

Say you run a small support bot that answers 50 questions a day. Assume each request sends about 1,500 tokens (the question, your instructions, a bit of history) and gets about 500 tokens back. Over 30 days that is 1,500 requests: 2.25M input tokens and 0.75M output tokens.

On Flash at off-peak prices, with no cache hits at all (the expensive case), the input costs 2.25 × $0.15 = $0.3375, the output 0.75 × $0.6 = $0.45, and the month comes to $0.3375 + $0.45 = $0.7875. If every single call landed in peak hours, everything doubles: 2 × $0.7875 = $1.575. On Pro at off-peak prices it is 2.25 × $0.66 = $1.485 for input and 0.75 × $1.98 = $1.485 for output, so $2.97 in total.

That is the whole “subscription question”: a flat plan would hide this arithmetic, a balance shows it. Double the questions, double the bill. Halve the answer length and the output half of the bill halves.

The same month on CheapTokens#

For the same bot, CheapTokens offers DeepSeek-V4-Flash (release 0731): one price at any hour of the day, so there is no peak calendar to plan around. The month above, 2.25M input and 0.75M output tokens, costs $0.109 there. With input and output counted together, that is cheaper than OpenRouter for this workload; the chart puts it next to the official prices from above:

Official prices: DeepSeek pricing page, checked 26 October 2026; no cache hits assumed.

If your bot is lighter, GLM-5.3-Flash is the other option: the same month costs $0.075, with a 390K context and answers up to 8,192 tokens. DeepSeek-V4-Flash gives you a 380K context and answers up to 32,768 tokens, which suits longer replies. Both sit behind one balance and one OpenAI/Anthropic-compatible API, so switching models is a one-line change. Current prices for every model are on the pricing page, and the DeepSeek page covers the model in more detail.

Try it: one request, then multiply#

The fastest way to get your own numbers is to send one real request and read the usage field of the response, which lists input and output tokens for that call. Replace YOUR_API_KEY with your key:

curl https://api.cheaptokens.dev/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek-ai/DeepSeek-V4-Flash-0731", "messages": [{"role": "user", "content": "Explain a prepaid balance in one sentence."}], "max_tokens": 100}'

Take the two token counts from usage, put them into the table above (or the official prices if you stay with DeepSeek), and multiply by the number of requests you expect per month. That is your budget. The full walkthrough of a first call is in your first API request, and the docs list every parameter.

Quick answers#

Is there a “Pro” plan? On the API, Pro is a model (deepseek-v4-pro), not a subscription tier. You pay its per-token prices from the same balance.

Why does my bill differ from the estimate? Usually history. Every request in a chat resends the earlier messages, so input grows with each turn; why a long chat costs more shows it step by step.

Do cache hits change the month? They can. On Flash off-peak, input that hits the cache costs $0.003 per 1M tokens against $0.15 for a miss (pricing page, 26 October 2026), so a bot that repeats the same long instructions in every request pays far less than the no-cache case above.