EN ▾
Get API key
https://api.deepseekapipricing.com/v1

DeepSeek API vs OpenAI API Cost: A Worked Comparison

A fair deepseek api vs openai api cost comparison needs your own token counts, not a headline price. This page gives you the formula, four worked workloads priced at $0.15 and $0.50 per 1M tokens, and a method for filling in any other provider's current rates.

Updated

Two balance scales weighing stacks of glowing tokens, illustrating a per-token API cost comparison

Key points

  1. Monthly cost = input tokens x input rate + output tokens x output rate; both sides matter.
  2. Here: $0.15 per 1M input, $0.50 per 1M output, no subscription, credit never expires.
  3. A 1,000-requests-a-day chatbot at 1,500 tokens in and 400 out costs $12.75 a month.
  4. Check official pricing pages for the other side; rates change, so this page does not quote them.

Short answer and a warning

Nobody can give you a trustworthy single number for how two APIs compare, because the answer depends on how many tokens you send, how many you get back, and how often a request fails or gets refused. What you can do is price your own workload on each side with the same formula. This page does that for the independent uncensored API sold on this site, and tells you where to look up the other side.

One clarification first. This service is not the official DeepSeek API and not the official OpenAI API. It runs one open-weight model, operated by an independent provider and tuned to answer without content refusals for lawful adult use. So the comparison here is between a different product and what you pay elsewhere, not a like-for-like price check of the same model. Rates for the official services change over time, so the honest instruction is to check their current pricing pages and use the formula below.

The formula

Every token-billed chat API follows the same shape:

cost = (input tokens / 1,000,000) x input rate + (output tokens / 1,000,000) x output rate

Here the input rate is $0.15 and the output rate is $0.50 per million tokens. Output costs a little over three times as much as input ($0.50 divided by $0.15 is 3.33). That ratio matters: a workload that reads a lot and writes a little is cheap, while one that writes long answers costs more per token.

If you want one blended number, pick a ratio. At three input tokens for every output token, the blended rate is (3 x $0.15 + $0.50) / 4 = $0.2375 per million tokens. At one input for every output, it is ($0.15 + $0.50) / 2 = $0.325. Use the blended number only for rough planning, and use the full formula for budgets.

Four workloads priced exactly

Each figure below follows from the two rates above. Check the arithmetic against your own counts by swapping in your numbers.

1. Support chatbot

1,000 requests a day for 30 days is 30,000 requests. Each carries 1,500 input tokens (system prompt plus history) and returns 400 tokens. Input: 30,000 x 1,500 = 45,000,000 tokens, which is 45 x $0.15 = $6.75. Output: 30,000 x 400 = 12,000,000 tokens, which is 12 x $0.50 = $6.00. Monthly total: $12.75, or $0.000425 per request.

2. Retrieval-augmented question answering

10,000 requests a month, each with 20,000 tokens of retrieved context and a 1,000-token answer. Input: 200,000,000 tokens = 200 x $0.15 = $30.00. Output: 10,000,000 tokens = 10 x $0.50 = $5.00. Monthly total: $35.00, or $0.0035 per request.

3. Batch summarization

100,000 documents, each 3,000 tokens in and a 300-token summary out. Input: 300,000,000 tokens = $45.00. Output: 30,000,000 tokens = $15.00. One-off total: $60.00, which is $0.0006 per document.

4. Long-context analysis

500 calls, each with 90,000 tokens of source text and a 2,000-token report. Input: 45,000,000 tokens = $6.75. Output: 1,000,000 tokens = $0.50. Total: $7.25, or $0.0145 per call.

WorkloadInput tokensOutput tokensCost hereSame workload elsewhere
Chatbot, per month45M12M$12.75fill in from their price page
RAG, per month200M10M$35.00fill in
Batch summaries, one-off300M30M$60.00fill in
Long context, 500 calls45M1M$7.25fill in

The last column is deliberately blank. Take the other provider's current input and output rates from their own site, apply the same formula, and you have a real comparison built from your numbers.

Costs that never show up in a price table

Two APIs with identical per-token prices can still cost different amounts. Check these before deciding.

  • Refusals and retries. If 10% of your calls are refused and you resend them rephrased, you make 1 / 0.9 = 1.11 calls per useful answer, an 11% overhead plus your developer time, as long as the provider bills refusals. Here, failed or refused requests are not charged, so the overhead is the time, not the tokens.
  • Answer length. Models differ in how wordy they are. If one model writes 600 tokens where another writes 400 for the same task, the output bill is 50% higher on the wordy one.
  • Context resend. In a chat, each turn resends the whole history as input. A 20-turn conversation does not cost 20 times a single turn; it costs closer to the sum of an arithmetic series.
  • Discounts and tiers. Some providers discount cached input, batch jobs or off-peak hours. If you qualify, include the discount in your estimate. The price here is flat.
  • Minimum spend and expiry. Prepaid credit here never expires and there is no subscription, so a quiet month costs nothing extra. Check whether others expire credit or require monthly commitments.

When price is the wrong deciding factor

A cheaper rate does not help if the model cannot do the job. Some honest cases where the other option is the better fit:

  • You need the official model from a specific vendor for contractual, compliance or capability reasons. This service runs a single independent open-weight model and cannot substitute for that.
  • You need image, audio or video generation, embeddings or fine-tuning. This API is text only.
  • Your workload depends on features that a vendor ships first in its own API. Check that the OpenAI-format features you use (streaming, tools, JSON mode) are all you need before switching.

The reverse also holds. If your pain is refusals on lawful adult content, such as fiction, roleplay or security research, a lower token price on a model that declines half your prompts is not a saving. In that case the useful number is cost per accepted answer, not cost per token.

A budgeting routine that holds up

Once you have the formula, a short routine keeps the estimate honest as your product changes.

  1. Measure a sample. Log prompt and completion tokens for a few hundred real requests. Do not guess from word counts; tokens run higher than words, and system prompts add up.
  2. Find the averages and the tail. The mean sets the budget, but a few very large requests set the worst day. If one in a hundred requests uses the full 100,000-token window, that is $0.015 of input for each of them, which is the same as the cost of about 35 ordinary chatbot requests at $0.000425.
  3. Add growth. Multiply by your expected traffic in three months, not today.
  4. Repeat the arithmetic for each option. Use the same token counts for each provider, and swap in only the rates and any discounts you actually qualify for.
  5. Add the cost of your own time. Moving a client to another endpoint takes an afternoon; building around refusals takes much longer.

With prepaid credit, you can also cap your risk by design. Credit is a hard ceiling: when it runs out, calls fail rather than running up a bill. Watching the X-Balance-USD header and topping up on a schedule is simpler than reconciling an invoice at the end of the month.

The same exercise works for a hobby project. Someone who runs a roleplay bot for friends with 200 messages a day at 2,000 tokens in and 300 out spends 200 x 30 = 6,000 requests a month: input 12,000,000 tokens ($1.80) plus output 1,800,000 tokens ($0.90), so $2.70 for the month. At that scale, the price per token barely matters, and quality and refusals decide the choice.

Test it with your own traffic

The surest way to settle the question is to measure. Every response includes a usage block, and the headers include X-Balance-USD, so you can read your remaining credit after each call. The snippet below runs a prompt and prints the exact cost from the returned token counts:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.deepseekapipricing.com/v1",
    api_key="YOUR_KEY",
)

IN_RATE = 0.15 / 1_000_000
OUT_RATE = 0.50 / 1_000_000

r = client.chat.completions.create(
    model="uncensored",
    messages=[{"role": "user", "content": "Summarize the plot of Hamlet in 120 words."}],
)
u = r.usage
cost = u.prompt_tokens * IN_RATE + u.completion_tokens * OUT_RATE
print(u.prompt_tokens, u.completion_tokens, f"${cost:.6f}")

Log those numbers for a day of real traffic, scale to a month, and compare with the same logic applied to your current provider's rates. New users can claim a $0.50 trial credit with one click on the API key page; it needs no card, is limited to one per person and lasts 7 days.

When you are ready to pay, top-ups run from $20 to $500 in whole dollars, with a 5% bonus from $50, 10% from $100 and 15% from $200. The chatbot above at $12.75 a month would make a $20 top-up last about 47 days, and a $100 top-up, which becomes $110 of credit, about 8.6 months. See the pricing page for payment methods, and the documentation for the full parameter list. If you came here through the DeepSeek V4 API page, note again that this is an independent service, not the vendor's own API.

01

Questions and answers

Is this the official DeepSeek or OpenAI API?

No. It is an independent service running one open-weight model, and it is not the official API of DeepSeek, OpenAI or any other vendor. Use their own sites for their prices.

How do I work out my monthly bill?

Multiply monthly input tokens by $0.15 per million and monthly output tokens by $0.50 per million, then add the two. The response usage block gives you the token counts.

Are refused or failed requests billed?

No. Failed or refused requests are not charged here. Check your other provider's terms on this point, because it changes the real cost of retries.

Does credit expire?

No. You pay as you go from prepaid credit, with no subscription and no expiry.

Your key is one form away

Create an account, copy the key, change the base URL. That is the whole setup.

Get API key