DeepSeek / Pricing
DeepSeek pricing:
chat access and API costs.
DeepSeek offers a free consumer assistant. Its developer API is a separate, paid service billed by token usage. Your API cost depends on the model, input caching, output length and peak or off-peak hours.
By China AI Chat Team · Sources and calculator checked 1 October 2026 · How we research
Is DeepSeek free?
The consumer assistant is available free. DeepSeek describes its official assistant as free in its Android app listing. To chat, use DeepSeek’s official website or follow its official app download links. Free access does not establish an unlimited-message or availability guarantee.
The official API is metered separately. Developers pay for the tokens processed by their applications. An API balance is not a chat-app subscription. DeepSeek’s published billing rules deduct usage from a topped-up or granted balance, using granted funds first when both are available. We do not assume that every new account receives promotional credit.
Other services that offer DeepSeek models set their own plans and prices. The rates below apply to DeepSeek’s own developer platform, not Together AI, OpenRouter or another host.
Official DeepSeek API prices
All rates are in USD per 1 million tokens. A token is a unit processed by the model, not a fixed number of words. Cached input, uncached input and generated output have separate rates.
| Model | Period | Cached input | Uncached input | Output |
|---|---|---|---|---|
| DeepSeek-V4.1-Flash | Off-peak | $0.003 | $0.15 | $0.60 |
| DeepSeek-V4.1-Flash | Peak | $0.006 | $0.30 | $1.20 |
| DeepSeek-V4-Pro-0813 | Off-peak | $0.022 | $0.66 | $1.98 |
| DeepSeek-V4-Pro-0813 | Peak | $0.044 | $1.32 | $3.96 |
The table and calculator use the rates recorded on 20 September 2026. We rechecked them against the official rate card on 1 October 2026; the prices and model mappings are unchanged.
deepseek-flash currently serves DeepSeek-V4.1-Flash; deepseek-v4-pro serves DeepSeek-V4-Pro-0813. See the DeepSeek model reference for identifiers and capabilities.
Choosing between the two IDs? Our Flash vs Pro API comparison reports actual token usage from six matched tasks and calculated peak/off-peak costs, including code calls that reported output-token usage without delivering final code. The examples below remain hypothetical workloads.
Peak hours: 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays. All other times, including weekends and Chinese public holidays, use off-peak prices. Off-peak rates are half the corresponding peak rates. Use UTC when scheduling work; your local clock may differ.
DeepSeek’s official pricing page is the source for these rates and the schedule. Prices and model mappings can change.
Estimate your API cost in three steps
- Choose the API model your application uses. Its ID must match the model selected in the calculator.
- Enter token counts for one representative request. If you already have a response, use its
usagefields below. If you have not made requests yet, use estimates and replace them with reported usage when available. - Enter the number of API requests you expect per month. Include follow-up calls and retries that generate billable usage. Use whole numbers without commas, then select Calculate cost.
| Calculator field | Value from the response |
|---|---|
| Uncached input tokens / request | usage.prompt_cache_miss_tokens |
| Cached input tokens / request | usage.prompt_cache_hit_tokens |
| Total output tokens / request | usage.completion_tokens, including reasoning |
If the cache split is unknown, enter the full input count (prompt_tokens, when reported) as uncached and set cached input to zero. Do not add reasoning_tokens again: it is already part of the completion total. The calculator compares the same workload entirely off-peak and entirely at peak; for a mixed schedule, calculate each group separately.
DeepSeek API cost calculator
Enter token counts for one request and the number of similar requests you expect in a month. Compare the same workload at off-peak and peak prices. Calculations run in your browser and make no AI API calls.
Use total completion tokens, including reasoning. Enter each input token once: cached or uncached. If your cache usage is unknown, put all input in uncached and leave cached at zero.
All off-peak
per month
All peak
per month
Two separate scenarios, not a price range guaranteed for your account. Each assumes identical token counts for every request and all requests in that period. For mixed workloads, calculate each group separately and add the totals. Estimates cover the token charges shown here, excluding any taxes, currency conversion or separately charged services.
Three worked cost examples
These are hypothetical workloads calculated from the rates above, not measured requests or invoices. Output counts include all generated completion tokens, including any reasoning tokens.
1. One short Flash request
Assume 1,000 uncached input tokens, zero cached input tokens and 500 output tokens.
Off-peak: (1,000 × $0.15 + 500 × $0.60) ÷ 1,000,000 = $0.00045 per request. The same token counts at peak rates cost $0.00090.
2. A monthly workload on Flash or Pro
Repeat that exact workload 1,000 times in a month. These are usage estimates, not monthly subscription plans.
| Model | All requests off-peak | All requests at peak |
|---|---|---|
| DeepSeek-V4.1-Flash | $0.45 | $0.90 |
| DeepSeek-V4-Pro-0813 | $1.65 | $3.30 |
For a mixed schedule, add the separately calculated groups. For example, 800 off-peak Flash requests and 200 peak requests with those token counts cost $0.54. Longer conversations or more generated tokens will change the result.
3. A longer input with a confirmed cache hit
Assume a Flash request contains 10,000 input tokens in total: 9,000 reported as cached and 1,000 uncached, plus 1,000 output tokens.
Off-peak: (9,000 × $0.003 + 1,000 × $0.15 + 1,000 × $0.60) ÷ 1,000,000 = $0.000777. At peak rates it costs $0.001554.
With all 10,000 input tokens uncached, the same output would instead cost $0.00210 off-peak or $0.00420 at peak. The saving applies only when the provider reports the cache hit; repeated text alone is not a guarantee.
What counts toward the bill?
Input includes the context you send
Count the system instructions, current message and conversation history included in each request. A short follow-up can still have a long input when your application sends earlier messages again. Use the API’s reported usage instead of estimating cost from the visible question alone.
Caching discounts eligible input
DeepSeek enables context caching by default, but a hit depends on matching a persisted prefix. The service describes caching as best effort. In the response’s usage object, prompt_cache_hit_tokens identifies cached input and prompt_cache_miss_tokens identifies uncached input. Together they make up prompt_tokens; do not add that total again. See the official caching guide.
Output can include reasoning
Use completion_tokens for the generated total. The API exposes reasoning as a breakdown within completion usage, so do not count it twice or estimate the bill from the final answer’s length alone. Thinking is enabled by default in the current API; see DeepSeek’s thinking-mode documentation.
max_tokens sets an output ceiling. It is not a fixed charge for that many tokens: the pricing rules bill actual usage. A lower ceiling can constrain output, but may also cut an answer short. The Chat Completions reference defines the limit and usage fields.
The cost formula
Cost in USD = [(cached input × cache-hit rate) + (uncached input × cache-miss rate) + (total output × output rate)] ÷ 1,000,000.
Apply the appropriate model and pricing period to each request. For a monthly budget, include follow-up calls and any retries that generate billable usage. Tool-driven workflows can involve several model calls for one user task. The calculator repeats the token counts you enter; it does not predict how much a model will generate.
Common DeepSeek pricing questions
How much does DeepSeek cost per month?
For the official API, there is no single monthly price in the published token rate card: the total comes from your usage. Choose the model, estimate each token category and multiply by your expected request count. The free consumer assistant is a separate service.
Is “DeepSeek Pro” a paid chat subscription?
In this price table, Pro means the DeepSeek-V4-Pro-0813 API model. Its token rate is not a monthly membership price for the chat app.
Why does another site show a different price?
Check the provider, model version, review date, cache assumption and pricing period. DeepSeek also says legacy deepseek-v4-flash and deepseek-v4-flash-vision-exp names now route to DeepSeek-V4.1-Flash and use Flash pricing. An old alias does not establish that you are buying the retired model at its old price.
How many messages can I send for $1?
There is no reliable fixed count. A message may produce a short reply, extensive reasoning or several tool calls, and later turns may resend earlier context. Use representative token usage and count API requests rather than treating every chat message as the same workload.
Choose your next step
For normal conversation, open the official DeepSeek chat. If you are building an application, continue to our DeepSeek API guide. For the wider product overview, return to DeepSeek explained.
This is an independent, documentation-based pricing guide. We rechecked DeepSeek’s pricing, token-usage, caching, thinking-mode and API reference pages, plus its official Android listing, on 1 October 2026. The token-usage guide explains why reported usage is more reliable than word or character estimates. We checked the arithmetic and calculator; we did not run paid DeepSeek API requests for this article. Before setting a budget, check the official rate card for changes to prices, model mappings and billing rules. Report a correction.