Software price guides AI DeepSeek
Last updated
DeepSeek pricing in the US
Hangzhou DeepSeek Artificial Intelligence sells DeepSeek to businesses in the United States only through its API, priced per million tokens in US dollars from a prepaid balance. The chat app and website are free. Peak-hour rates are the list price, and off-peak hours cost half.
Country
- Starting price
- Pay as you go
- API, per million tokens, in US dollars
- Pricing model
- Pay as you go, per million tokens
- Half price off-peak
- Billing
- Prepaid balance
- Top-ups do not expire
- Tax
- Tax may be added at checkout
- All prices in USD
- YOY change
- Not enough history
- Last updated
- 18 Oct 2026
How much does DeepSeek cost in the United States?
Plans
One pay-as-you-go API, billed in US dollars from a prepaid balance.
| Plan | Monthly | Who it is for | Key limits |
|---|---|---|---|
| DeepSeek API | Pay per use | Pay as you go, per million tokens | Key limits
|
Additional costs
Costs on top of the token price.
| Cost | Price | Source |
|---|---|---|
| Peak-hour usage | Twice the off-peak rateWeekdays 01:00 to 04:00 and 06:00 to 10:00 UTC, excluding Chinese public holidays | |
| deepseek-v4-pro in place of deepseek-flash | US$1.32 input; US$3.96 outputPer million tokens, at peak | |
| Cache-miss input | 50 times the cache-hit rate on deepseek-flashInput not served from the automatic cache | |
| Image input on deepseek-flash | The model's input rateBilled as input tokens, up to 384 tokens an image | |
| Files API | FreeUpload once, reference by file id | |
| Search and Fetch APIs | US$0.75Per 1,000 successful requests each; free through DeepSeek Harness, Claude Code or Codex until 1 November 2026 | |
| Higher concurrency limit | No additional costOn request, through a capacity expansion form | |
| Running out of balance | Requests stopThe API returns error 402 until the account is topped up | |
| Sales tax | Not publicly listedDeepSeek says nothing about US sales tax |
API prices
The DeepSeek API is billed by use, in US dollars per million tokens, at the peak rate; off-peak is half.
| Model | Input | Output | Cached input | Context window | Source |
|---|---|---|---|---|---|
| DeepSeek V4 Pro | US$1.32 | US$3.96 | US$0.04 | 1M tokens | |
| DeepSeek V4.1 Flash | US$0.30 | US$1.20 | US$0.01 | 1M tokens |
DeepSeek cost calculator
Enter your monthly API use and pick a model. Estimates in USD at peak rates, before any sales tax; off-peak use costs half.
What jobs do you want AI to do?
How many times a month for each job. For example, 300 if you sign about 300 contracts a month.
Writing
- Answer support emailsa month
- Summarise email threadsa month
- Answer questions from your documentsa month
- Summarise contractsa month
- Draft blog posts or newslettersa month
- Review code changesa month
Decisions
- Route support ticketsa month
- Score inbound leadsa month
- Flag invoices for reviewa month
- Check documents against a policya month
Estimated cost
US$0.41a month
- Per year, billed monthly
- US$4.92
- Automated jobs
- US$0.41
How we worked it out
- Answer support emails: about 1,100 words read and 200 written each time, 500 times a month.
Estimated, USD, excluding sales tax.
DeepSeek discount codes and offers
Current DeepSeek offers in the United States, checked 18 October 2026.
-
Off-peak pricing
50% off peak rates on all billing items
Applied automatically
Requests outside 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday; weekends and Chinese public holidays are off-peak all day
How to negotiate DeepSeek's price
Ranked by confidence of the biggest return.
| Rank | Lever | How it works | What it can save |
|---|---|---|---|
| [01] | Run batch work off-peak | Off-peak is half the peak rate. Peak is 9pm to midnight Sunday to Thursday and 2am to 6am Monday to Friday, US Eastern daylight time; an hour earlier from 1 November. Weekends are mostly off-peak. | 50% of usage spend |
| [02] | Let repeated context hit the cache | Caching is automatic. Cache hits cost 2% of the cache-miss input rate on deepseek-flash and about 3.3% on deepseek-v4-pro. | Up to 98% of input cost on cached tokens |
| [03] | Use deepseek-flash before deepseek-v4-pro | At peak, Flash is US$0.30 in and US$1.20 out per million tokens; V4-Pro is US$1.32 and US$3.96. | About 77% of input and 70% of output cost |
| [04] | Get unused balance refunded | Top-ups do not expire, and unused balances can be refunded online. | Any balance left unspent |
| [05] | Ask for higher limits instead of paying more | There are no tiered plans. DeepSeek grants higher concurrency on request at no extra cost. | No fee for higher limits |
DeepSeek price history
DeepSeek's API list prices over the last two years, in US dollars per million tokens, from DeepSeek's own announcements and archived pricing pages.
API prices
DeepSeek's API price over the same two years, one line per model family, in US dollars per million tokens. Each point is the blended price, 70% input and 30% output; hover for each launch and change.
DeepSeek alternatives
Other tools in our price guides for the US, largest first by market share.
| Vendor | Starting price | Pricing model | Key difference | |
|---|---|---|---|---|
| ChatGPT | Free plan; paid from US$8 a user a month (ChatGPT Go) | Per person for one user, per seat for teams, plus usage credits | Per person or per seat in USD, sales tax extra, plus usage credits | Price guides |
| Gemini | Free plan; paid from US$4.99 a user a month (Google AI Plus) | Per person on Google AI plans, per user in Google Workspace, plus AI credit top-ups | Per person on Google AI plans, or per user inside Google Workspace, in USD | Price guides |
| Claude | Free plan; paid from US$20 a user a month (Claude Pro) | Per person for one user, per seat for teams, plus usage at API rates | Per person or per seat in USD before sales tax, plus usage at API rates | Price guides |
| Meta AI | Free plan; paid from US$7.99 a user a month (Meta One Core) | Free for everyday use, per person for Core and Premium, per profile for business plans, plus API usage | Free assistant, with paid Meta One bundles per person or per business profile | Price guides |
| Grok | Free plan; paid from US$8 a user a month (Grok Plus) | Per person for one user, per seat for teams, plus prepaid usage credit | Per person or per seat in USD before sales tax, plus prepaid usage credit | Price guides |
| Mistral Vibe | Free plan; paid from US$15 a user a month (Mistral Vibe Pro) | Per person for one user, per seat for teams, plus usage at API rates | Billed in US dollars | Price guides |
| Qwen | US$8 a user a month (Qwen Token Plan Lite) | Per person for one developer, per seat for teams, plus API usage per token | Per person or per seat in US dollars on monthly Credits quotas, plus API usage per token | Price guides |
| Kimi | Free plan; paid from US$19 a user a month (Kimi Plus) | Per person for one user, per seat for teams, paid yearly, plus usage top-ups | Per person or per seat in US dollars, Business paid yearly, plus top-ups at about API rates | Price guides |
| Z.ai | Free plan; paid from US$18 a user a month (GLM Coding Plan Lite) | Per person for one user, per seat for teams, plus Team overage at API rates | Per person or per seat in US dollars before tax, 20% off quarterly and 30% off yearly | Price guides |
| MiniMax | Free plan; paid from US$22 a user a month (MiniMax M Plan Go) | Per person for one user, no team plan, plus credit packs past plan limits | Per person in US dollars before tax, no team plan, plus credit packs past plan limits | Price guides |
| Jev | Pay as you go API, per million tokens | Pay as you go, per input token, output free | API only, per billion input tokens in US dollars, output free | Price guides |
DeepSeek pricing questions
The DeepSeek chat app and website are free. The API is pay as you go in US dollars. At peak, deepseek-flash costs US$0.30 per million input tokens and US$1.20 per million output tokens, and deepseek-v4-pro costs US$1.32 and US$3.96. Off-peak use is half those rates.
Peak is 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. On Eastern daylight time, until 1 November 2026, that is 9pm to midnight Sunday to Thursday and 2am to 6am Monday to Friday; on Pacific daylight time, 6pm to 9pm Sunday to Thursday and 11pm to 3am overnight into Monday to Friday. On standard time each is an hour earlier. The US working day is off-peak. This conversion is Ensurva's, not DeepSeek's.
DeepSeek does not say. Its pricing page and terms list prices in US dollars with no mention of sales tax, so prices here are shown before any tax. Rates vary by state.
Top up a prepaid balance by PayPal, bank card, Alipay or WeChat Pay. Usage is deducted per token, any granted free balance first. Top-ups do not expire and unused balances can be refunded. When the balance runs out, the API returns error 402 until it is topped up.
No. DeepSeek has one pricing standard with no tiered plans, and grants higher rate limits on request at no extra cost. The discounts are off-peak hours, at half price, and cached input, at 2% of the cache-miss rate on deepseek-flash.
Several times. DeepSeek V3.2 cut prices on 29 September 2025 to US$0.28 input and US$0.42 output. V4 launched on 24 April 2026, with Flash at US$0.14 and US$0.28. From 16 August 2026 DeepSeek raised V4 prices and brought in peak and off-peak billing, Flash to US$0.44 and US$1.32 at peak. V4.1 Flash cut that on 10 September 2026 to US$0.30 and US$1.20.
DeepSeek pricing sources and methodology
From DeepSeek's English pricing page, FAQ, API documentation and platform terms, read on 18 October 2026. DeepSeek publishes one price list for every country and no list of countries it serves, so these are the same prices DeepSeek shows everywhere. DeepSeek sells only its API; the chat and apps are free, and no paid subscription was found on DeepSeek's own pages. Prices are the peak rates, which are the list price; off-peak is half. DeepSeek bills accounts in US dollars and says nothing about sales tax, so amounts are shown before any tax.
Estimates at peak rates, with input priced at the cache-miss rate, on the chosen model from Ensurva's AI price guides, in US dollars. Cached and off-peak use cost less than the estimate. There is no company size table, because DeepSeek is priced by use rather than by seat.
DeepSeek states peak hours in UTC. The US times are Ensurva's conversion, on daylight time until 1 November 2026 and standard time after it, for the Eastern and Pacific time zones.
The API table comes from the same data as Ensurva's AI price guides, read from DeepSeek's own pricing page, in US dollars per million tokens at the peak rate.
DeepSeek's release announcements and archived copies of its pricing page from November 2024 to October 2026, in US dollars per million tokens, the same in every country. Peak rates are shown from August 2026, when peak and off-peak billing began.
Drawn from how DeepSeek charges, not from records of negotiated deals.
Not collected for the United States yet, so the section is left off the page.
Checked monthly against DeepSeek's own pages.
The calculator's By use case tab puts each person on a seat that suits how heavily their role uses AI. Software developers use it most, about 15% of their working hours against about 6% across all workers (Federal Reserve Bank of St. Louis, 2026), so they get the heavier seat where the plan has one. Content, support and finance staff fit a standard seat, and general office staff use AI least. Each automated job is priced at a typical size, about 100 tokens to 75 words: a support reply reads about 1,100 words and writes 200, and a contract summary reads 10,000 and writes 750. Decision jobs read the record and give a short typed answer: about 600 words to route a support ticket, 4,500 to check a document against a policy. Those sizes are Ensurva's own estimates. A model that writes no text, such as Jev, is priced on decision jobs only. Jobs are priced as API use on the vendor's default model, less any API credits the plan includes.
- DeepSeek Models and Pricing 18 Oct 2026
- DeepSeek English FAQ, API 18 Oct 2026
- DeepSeek Search and Fetch 18 Oct 2026
- DeepSeek Open Platform terms 18 Oct 2026
- DeepSeek change log 18 Oct 2026
- DeepSeek Get User Balance API 18 Oct 2026
- DeepSeek error codes 18 Oct 2026
- DeepSeek rate limit 18 Oct 2026
- DeepSeek V4.1-Flash release 18 Oct 2026
- DeepSeek V4-Pro and new pricing 18 Oct 2026
- DeepSeek V4-Flash-Vision-Exp release 18 Oct 2026
- DeepSeek V4 preview release 18 Oct 2026
- DeepSeek V3.2-Exp release 18 Oct 2026
- DeepSeek-V3 release 18 Oct 2026
- DeepSeek pricing page, archived November 2024 16 Oct 2026
- DeepSeek pricing page, archived September 2025 16 Oct 2026
- DeepSeek pricing page, archived June 2026 16 Oct 2026
Prices are estimates from DeepSeek's published US dollar list prices on the date shown. Check api-docs.deepseek.com before buying.