Last Updated: October 12, 2026

DeepSeek pricing has no subscription. The chat app is free, and the API is pay-as-you-go. DeepSeek V4.1 Flash costs $0.30 per million input tokens and $1.20 per million output tokens at peak, and V4 Pro costs $1.32 and $3.96. Off-peak hours are half price, and cached input is up to 98% cheaper. Prices come from DeepSeek’s API docs, checked October 2026.

DeepSeek is one of the cheapest ways to run a frontier-class model, but three price lists are in circulation and they do not agree. DeepSeek’s own docs page, its marketing page and the third-party guides all show different numbers for the same models.

This guide uses the current docs page, shows what each rate means in dollars per request, and explains the off-peak discount and the cache discount that can cut a bill by half or more. If you want the background on the company and its models first, start with our guide to what DeepSeek is.

💡 Not sure which AI tool is actually right for your business?

Get the free guide, Which AI Tool Should You Actually Use? — a straightforward breakdown of the leading AI tools to help you pick the right one for your needs.

Subscribe to AI Business Weekly for the guide, plus daily coverage of AI trends, acquisitions, and product launches.

Get the Free Guide →

DeepSeek costs nothing to chat with and is billed by usage through the API. At peak hours, V4.1 Flash costs $0.30 per million input tokens and $1.20 per million output tokens. V4 Pro costs $1.32 and $3.96. Off-peak rates are half those prices. DeepSeek lists no subscription plan.

Source: DeepSeek’s Models & Pricing page, checked October 2026. DeepSeek reserves the right to change prices and recommends checking the page regularly. The page’s table does not label which column belongs to which model, so I matched the columns by order and cross-checked them against BenchLM and AI Pricing Guru, which list the same figures.

Yes, for the chat app and website. CloudZero reports that the chat interface at chat.deepseek.com is free to use, though DeepSeek’s own platform page does not state it. The API is not free: there is no permanent free tier, and new accounts only sometimes receive a promotional balance.

DeepSeek’s docs mention a “granted balance” that is spent before your topped-up balance, but they give no amount, eligibility rules or expiry. Treat any free API credit as a bonus, not a plan. The models are also open-weight under the MIT license, according to CloudZero, so you can run them yourself on your own hardware, though that carries its own GPU cost.

For other no-cost options, see our guide to the best free AI tools.

Sources: DeepSeek’s pricing documentation, CloudZero’s DeepSeek pricing guide (updated September 2026) and AI Pricing Guru.

The DeepSeek API charges per million tokens, split into cache-hit input, cache-miss input and output. V4.1 Flash is the cheaper model, at $0.30 input and $1.20 output at peak. V4 Pro costs about four times more for input and three times more for output. Both support a 1 million token context.

All prices are in US dollars per million tokens. The multiples are my arithmetic: $1.32 divided by $0.30 is 4.4 for input, and $3.96 divided by $1.20 is 3.3 for output.

Cache hits matter most for apps that send the same long prompt repeatedly. Cached input on Flash costs $0.006, which is 2% of the $0.30 cache-miss price, so a cache hit is 98% cheaper. On Pro the cut is about 97%. Caching is automatic on repeated prefixes, according to AI Pricing Guru. Fees come out of your topped-up balance, or out of any granted balance first.

Source: DeepSeek’s Models & Pricing page. For how this compares with other APIs, see our Claude API pricing guide and ChatGPT pricing guide.

DeepSeek charges half price outside peak hours. Peak hours are 01:00 to 04:00 UTC and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. Every other hour is off-peak, including all of the weekend. The discount applies to cache-hit, cache-miss and output rates alike.

In North American time, 01:00 to 04:00 UTC is 9 PM to midnight Eastern, and 06:00 to 10:00 UTC is 2 AM to 6 AM Eastern, during daylight saving time. After clocks change in November, the windows shift one hour earlier on the clock. That conversion is mine, so check it against your own time zone before scheduling jobs.

The practical effect is large for batch work. A job that runs on a Saturday costs half what the same job costs at 2 AM Eastern on a Tuesday. Interactive apps cannot choose their hours, so the discount mostly helps scheduled and background workloads. DeepSeek’s page does not say how it verifies Chinese public holidays for billing.

Source: footnote 2 of DeepSeek’s Models & Pricing page.

A typical request with 1,000 input tokens and 500 output tokens costs about $0.0009 on V4.1 Flash at peak, or roughly 1,100 requests per dollar. The same request costs about $0.0033 on V4 Pro, or about 300 requests per dollar. Off-peak rates cut each figure in half.

These estimates are my arithmetic from DeepSeek’s published rates, assuming every input token is a cache miss:

Real costs depend on answer length and how much of your prompt is cached. DeepSeek enables thinking mode by default on both models, and a longer reasoned answer means more output tokens, so a thinking response can cost more than the table suggests. Run a sample of your own prompts and read the usage numbers before you forecast a budget.

Use V4.1 Flash for most work: it is the cheaper model, it supports vision and it has the higher concurrency limit of 2,500. Use V4 Pro when answer quality on hard reasoning or coding tasks justifies about four times the input price. Test both on your own prompts.

Two cautions apply before you build on the API. CloudZero notes that DeepSeek runs from China-based infrastructure, which can affect latency for US and European users and matters if you have data residency rules. And BenchLM reports that requests above your account’s concurrency limit get an HTTP 429 error, regardless of how many API keys you use. For a comparison with the best-known chatbot, see our DeepSeek vs ChatGPT guide, and for adoption numbers see DeepSeek statistics.

DeepSeek’s prices differ because the company has renamed and retired models several times, and its own pages are not in sync. The docs page lists V4.1 Flash at $0.30 input and $1.20 output at peak. DeepSeek’s platform page lists V4 Flash at $0.14 and $0.28. CloudZero lists V4 Pro at $1.74 and $3.48.

The older figures belong to earlier versions or promotions. According to BenchLM, deepseek-v4-flash was retired on September 10, 2026 and now routes to V4.1 Flash at the Flash price, and deepseek-chat and deepseek-reasoner reached their retirement deadline after July 24, 2026. DeepSeek’s docs page confirms the legacy Flash names are retired and billed at the Flash price, and its quick-start guide lists deepseek-flash and deepseek-v4-pro as the current model names. Neither page mentions deepseek-chat or deepseek-reasoner.

The safest approach is to treat the docs pricing page as the source of truth and re-check it before you commit to a budget. If you are comparing against a guide that lists $0.14 for input or $0.28 for output, it is almost certainly describing an older model.

Sources: DeepSeek’s docs, DeepSeek’s platform page, CloudZero and BenchLM.

DeepSeek pricing is free in the chat app and pay-as-you-go on the API, as of October 2026. V4.1 Flash costs $0.30 per million input tokens and $1.20 per million output tokens at peak, V4 Pro costs $1.32 and $3.96, and off-peak hours halve every rate. A typical request costs about $0.0009 on Flash, and cache hits cut input costs by 96% to 98%. Many guides list older prices, so use DeepSeek’s docs page as your reference. Re-check it before budgeting, because DeepSeek says it can change prices at any time.

The DeepSeek chat app is free, and the API is pay-as-you-go. V4.1 Flash costs $0.30 per million input tokens and $1.20 per million output tokens at peak, as of October 2026. V4 Pro costs $1.32 and $3.96. Off-peak hours are half price, and DeepSeek lists no subscription plan.

The chat app and website are free, according to CloudZero. The API has no permanent free tier, though new accounts sometimes get a promotional balance, and DeepSeek does not publish the amount. The models are also open-weight under the MIT license, so you can self-host them if you cover the hardware cost.

DeepSeek bills per million tokens. V4.1 Flash is $0.006 for cached input, $0.30 for uncached input and $1.20 for output at peak. V4 Pro is $0.044, $1.32 and $3.96. Off-peak rates are half those figures. A request with 1,000 input and 500 output tokens costs about $0.0009 on Flash.

Peak hours are 01:00 to 04:00 UTC and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. All other times are off-peak, including weekends, and cost half as much. In daylight saving time, that is 9 PM to midnight and 2 AM to 6 AM Eastern.

No subscription appears on the DeepSeek pages I checked. The chat app is free, and the API runs on a prepaid balance that is deducted as you use tokens. Fees come from your topped-up balance, or from any granted balance first. Check the platform for current top-up terms.

DeepSeek has renamed and retired models and run promotions, so older guides describe earlier prices. Some pages list V4 Flash at $0.14 input and $0.28 output, while the current docs page lists V4.1 Flash at $0.30 and $1.20 at peak. Use the docs pricing page as the reference.

By Sameer Khan

AI-assisted. Researched, reviewed, and edited by Sameer Khan before publication