DeepSeek API vs OpenAI API Cost: Worked Examples

Why headline prices mislead

Anyone asking about DeepSeek API vs OpenAI API cost is usually looking at two price tables with different models, different caching rules and different output multipliers. A table shows a rate per million tokens; your bill depends on how many of each kind of token you send. A chatbot that resends long histories is input-heavy. A story generator is output-heavy. The same two price lists can rank in opposite order for those two workloads.

So skip the table-reading and do the arithmetic on your own traffic. The method takes ten minutes and works for any provider, including the one this site sells. We do not quote other vendors' prices here, because they change; where a number from them is needed, the text says to check their current pricing.

The formula

Write it down once:

Monthly cost = (input tokens in millions × input price per 1M) + (output tokens in millions × output price per 1M)

Here the prices are $0.15 per 1M input tokens and $0.50 per 1M output tokens. Output costs 3.33 times as much per token as input, so a workload's output share decides much of the bill. Billing is pay as you go from prepaid credit, with no subscription. Failed or refused requests are not charged, and credit never expires.

For another provider, find three things on their pricing page: the input price, the output price, and whether any discount applies to repeated prompt prefixes. Then apply the same formula to the same token counts. Different models may need different numbers of tokens to finish the same task, so compare cost per finished task as well as cost per token.

Five worked monthly examples at our rates

Where do such token counts come from? Take the chatbot row. If each conversation has 8 turns and the whole history is resent every turn, a conversation might use about 4,000 input tokens and 800 output tokens. Ten thousand conversations a month is 40M input and 8M output, which is the row above. Halve the history by trimming old turns and the input line falls by half, to $3.00, which brings the month to $7.00. History management is often a bigger saving than switching vendors.

Notice the shape. For input-heavy jobs the input line dominates; for the fiction generator the output line is nearly nine tenths of the bill. If you are comparing vendors, the question is which of the two prices matters for you. A vendor with a low input price but a high output price looks great for summarisation and poor for writing.

Get your real token counts first

Guessing token counts is the weakest part of any cost comparison. Measure them. The API returns a usage object with every response, so a short script can sample real traffic and project a month.

Copyfrom openai import OpenAI

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

PRICE_IN, PRICE_OUT = 0.15, 0.50  # dollars per 1M tokens
samples = ["Summarise: ...", "Draft a reply to: ...", "Classify: ..."]

tin = tout = 0
for s in samples:
    r = client.chat.completions.create(
        model="uncensored",
        messages=[{"role": "user", "content": s}],
    )
    tin += r.usage.prompt_tokens
    tout += r.usage.completion_tokens

cost = tin / 1e6 * PRICE_IN + tout / 1e6 * PRICE_OUT
print(f"{tin} in, {tout} out, ${cost:.6f} for {len(samples)} requests")
print(f"projected monthly at 100k requests: ${cost / len(samples) * 100000:.2f}")

Replace the sample prompts with 20 to 50 real ones and the projection becomes useful. To compare another vendor, change the two price constants to theirs, using the numbers on their current pricing page. Token counts can differ slightly between tokenizers, so treat the other side's projection as an estimate.

What else changes the bill

• Prompt caching and discounts. Some providers discount repeated prefixes. Check whether each vendor does, and how.
• Reasoning tokens. A model that thinks before answering may bill extra output tokens. Compare total tokens per finished task.
• Retries and failures. Here, failed or refused requests are not charged. Check the equivalent policy elsewhere.
• Minimum spend. Top-ups here 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.
• Rate limits. A full key allows 8 concurrent requests and 300 per minute. A batch that needs more parallelism will take longer than the price alone suggests.

What cheaper does not mean

Also weigh the non-price factors that matter to your project. Content policy is the main one for many readers of this page: a model tuned to answer without refusals for lawful adult use behaves differently from a general assistant, which is a benefit for fiction and roleplay products and a point to manage for anything customer-facing. There is one hard limit, no sexual content involving minors. Data handling matters too, and here prompts are not used for training.

Cost is only half the comparison. The model behind this API is a single open-weight model, run by an independent provider and tuned to answer without content refusals for lawful adult use. It is not the official API of DeepSeek or OpenAI, and it is not a router across vendors. Its context window is 100,000 tokens, it is text only, and it has no embeddings or fine-tuning. If your task needs a specific vendor's flagship model, the cheaper token price buys you nothing.

The fair test is task quality per dollar. Take 30 of your real prompts, run them through each candidate, score the answers blind, and divide the bill by the number of acceptable answers. Sometimes a cheap model needs two attempts and loses its edge; sometimes it matches the expensive one and you save most of the spend.

Switching in practice

If your code already uses an OpenAI SDK, trying this endpoint is a three-line change: base URL, key and model id uncensored. The Python walkthrough shows the full pattern, and the DeepSeek API key article explains where keys fit. New accounts can claim $0.50 of trial credit with one click on the API key page. That is enough to run the measurement script above many times at no cost. Rates and bonuses are on the pricing page, and parameters are in the docs. You can pay with card (Visa, Mastercard, Apple Pay, Google Pay) or crypto (USDT on TRC20, USDC on Base).

More guides

• DeepSeek API SDK: Use the OpenAI SDK and Swap Endpoints
• DeepSeek API Key: Uncensored LLM API Access