A necessary clarification
This site is an independent, OpenAI-compatible API. It is not DeepSeek's official API and does not resell it. It runs a single open-weight model hosted by an independent provider, tuned to answer without content refusals for lawful adult use. The page exists for people who searched for DeepSeek pricing and want to compare it with another per-million rate card.
Because the rates of other providers change, no DeepSeek price is quoted here. Open their pricing page, copy the current numbers for input and output, and plug them into the formula below. Doing that takes two minutes and avoids relying on a figure that went stale.
How per-million pricing works
Language model APIs bill by tokens, a token being a short chunk of text (a rough English average is about four characters). Rates are quoted per one million tokens, with two separate lines:
- Input tokens: everything you send, including the system prompt, the history and any documents.
- Output tokens: everything the model writes back.
Output usually costs more per token than input, so the shape of your workload matters. A summariser reads a lot and writes a little. A story generator does the opposite. Two workloads with the same total token count can differ in cost by a factor of two or more.
Some providers also price cached input differently or add off-peak discounts. When you compare, check whether the rate you copied is the standard one or a special one, and compare standard to standard.
The formula:
cost = (input tokens / 1,000,000) x input rate + (output tokens / 1,000,000) x output rate
This API's rate card
| Item | Value |
|---|---|
| Input | $0.15 per 1M tokens |
| Output | $0.50 per 1M tokens |
| Billing | Prepaid credit, pay as you go, no subscription |
| Failed or refused requests | Not charged |
| Credit expiry | Never |
| Context window | 100,000 tokens (prompt plus completion) |
Top-ups are any whole amount from $20 to $500, with +5% from $50, +10% from $100 and +15% from $200. A $200 top-up therefore gives $230 of credit. Pay by crypto (USDT on TRC20, USDC on Base). The pricing page has the full list.
Monthly cost examples
Four workloads, computed from the two rates above.
1. Small support bot: 10,000 requests a month
1,000 input and 400 output tokens each. Input: 10M tokens x $0.15 = $1.50. Output: 4M x $0.50 = $2.00. Total: $3.50.
2. Busy chat app: 200,000 requests a month
1,500 input and 300 output. Input: 300M x $0.15 = $45. Output: 60M x $0.50 = $30. Total: $75.
3. Document pipeline: 50,000 documents a month
6,000 input and 400 output. Input: 300M x $0.15 = $45. Output: 20M x $0.50 = $10. Total: $55.
4. Heavy generation: 2,000,000 short requests a month
500 input and 200 output. Input: 1,000M x $0.15 = $150. Output: 400M x $0.50 = $200. Total: $350.
Note how workload 4 spends more on output despite using fewer output tokens. That is the effect of the higher output rate, and it is why the mix matters more than the total volume.
A calculator you can run with any rates
To compare with any provider, put their numbers next to ours. Replace the second rate card with the values from their pricing page.
def monthly(requests, tin, tout, rate_in, rate_out):
# rates are dollars per 1M tokens
return requests * (tin * rate_in + tout * rate_out) / 1_000_000
ours = (0.15, 0.50)
theirs = (0.0, 0.0) # fill in from the other provider's current pricing page
for name, (ri, ro) in {"this API": ours, "other": theirs}.items():
print(name, round(monthly(200_000, 1500, 300, ri, ro), 2))
Use real numbers for tin and tout. Run a few real requests and read the usage object returned with each response, then average. To test with this API at the base URL https://api.deepseekapipricing.com/v1, send your request with model uncensored:
curl https://api.deepseekapipricing.com/v1/chat/completions \
-H "Authorization: Bearer $API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "uncensored", "messages": [{"role": "user", "content": "Say hello."}]}'Estimating tokens before you have data
If you have not run anything yet, use rough conversions. English text averages about four characters per token, and 1,000 tokens is roughly 750 words. Code and non-English text usually take more tokens per word, so add a margin.
Example: generating a 1,500-word article from a 300-token brief. Output is about 2,000 tokens. Input: 300 / 1,000,000 x $0.15 = $0.000045. Output: 2,000 / 1,000,000 x $0.50 = $0.001. One article costs about $0.001045, so 1,000 articles cost about $1.05.
Once you have real traffic, stop estimating. Every response carries an X-Balance-USD header, so you can log the balance before and after a batch and divide the difference by the request count. That is your true cost per request, with no assumptions.
What price does not tell you
A lower rate card is only a bargain if the model does your job. Before you move a workload, weigh these questions:
- Does the output quality hold up on your prompts? Test fifty real ones, not a hello-world.
- Do you need refusals handled differently? This model is tuned to answer without content refusals for lawful adult use. If you need a particular vendor's model by name, a stand-in is not a replacement.
- Do you need features this API lacks? It is text only, with no embeddings, images or fine-tuning.
- Are the limits enough? A full key gets 8 concurrent requests and 300 per minute.
For side-by-side cost work on a related pair, see DeepSeek API vs OpenAI API cost, and if you are testing before paying, the free trial page explains the $0.50 starting credit. The docs cover the request format.

