Is DeepSeek free? The app is, and the API starts at $0.15 per million input tokens

In brief

Yes, DeepSeek's chat is free: the website and the iOS and Android apps have no paid tier and no in-app purchases as of October 2026. The API is not free. It bills per token, and V4.1 Flash costs $0.15 per million uncached input tokens off-peak, which is 79% of the week. The open weights are free under the MIT license.

A blue glass whale resting on a phone beside a blank price tag, a brass coin meter and hourglass, and a black server box, each on its own black plinth in a dark studio.

Is DeepSeek free to use? The chat is, the API is not

DeepSeek has four ways in, and two of them cost nothing. The chat at chat.deepseek.com and the phone apps are free. The API charges for every token. The open weights are free to download, but you pay for the computer that runs them.

1,000 DeepSeek chat turns cost $0 in the app and $0.45 to $3.30 on the API
off-peakweekday peak$0$1$2$3$4per 1,000 turnsDeepSeek app andwebsite$0$0no paid tierAPI, V4.1 Flash$0.45$0.90deepseek-flash in the APIAPI, V4 Pro$1.65$3.30Expert Mode in the app

A turn here is 1,000 tokens in and 500 out, priced from DeepSeek's own pricing page, published 24 September 2026. That's a short question and a medium answer. Thinking mode is on by default in the API, and the reasoning it writes is billed as output, so hard questions cost more than this.

RoutePrice, October 2026What you getThe catch
Website, chat.deepseek.com$0Chat, Search, file and image uploads, Expert Mode on V4 ProServer busy refusals at peak, chats stored in China
App, iOS and Android$0, no in-app purchases listedThe same account and chat history as the websiteThe same limits as the website, rated 13+
DeepSeek API$0.15 to $0.30 per 1M uncached input tokens on V4.1 FlashCode access to V4.1 Flash and V4 Pro, 1M token contextPrepaid balance, calls stop at zero
Open weights on Hugging Face$0, MIT licenseV4.1 Flash, 552B parameters510 GB of weight files, server hardware
R1 distills through Ollama$01.5B to 70B models for a laptop or desktopAn older, much smaller model than the app runs
OpenRouter and other hostsPer token, no free DeepSeek ids on 12 October 2026The same models on other serversEach host has its own privacy policy

The App Store's own listing, read on 12 October 2026, opens with "Experience seamless interaction with DeepSeek's official AI assistant for free!" and shows a price of Free with no in-app purchases. The listing had 12,000 ratings at 3.9 stars that day.

Is DeepSeek free and unlimited?

DeepSeek is free without a published cap, but it isn't unlimited in practice. Its terms of use, its help centre FAQ and its App Store listing, all read on 12 October 2026, name no daily message limit. What stops you is capacity.

When DeepSeek's servers are full, the chat refuses with a "server is busy" message. Its API docs file the same thing as error 503, "Server Overloaded", caused by "high traffic". Retrying a minute later often works. When the message names the search service, turning Search off helps, and our server busy guide walks through the order that works.

Two other ceilings show up in DeepSeek's own FAQ. Uploads can fail with "Exceeded maximum upload limit." Very long chats can fail to load because "The chat context is too long." Neither comes with a number. Start a new chat when a thread gets long, and you'll hit both less.

There's also no way to pay your way past the queue. DeepSeek sells no paid chat tier, so nobody gets priority. Its API pricing marks the busy hours instead: prices double from 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, except on Chinese public holidays. That's 35 of the week's 168 hours. If the chat keeps refusing you, try outside them.

What does the free DeepSeek chat leave out?

The free chat gives you text, not media. DeepSeek's terms describe what the service returns as "text, tables, and code", so the app can read an image you upload but won't make one. There's no image, video or music generation anywhere in it.

It also leaves out three things people often assume come with a free account:

  • No API calls. The chat account doesn't come with free API use. The API runs on a separate prepaid balance and returns error 402, "Insufficient Balance", once it hits zero.
  • No promise it's up. The terms supply the service "as is" and "as available", and with no paid tier there's no priority lane to buy.
  • No private default. DeepSeek's privacy policy, last updated 10 February 2026, says it uses your chats to train its models unless you opt out.

That last one is the real price of the free chat. In the policy's own words, "we directly collect, process and store your Personal Data in People's Republic of China." It lists a right to opt out of training, but the storage location doesn't change. Is DeepSeek safe covers what that means for work data, and deleted chats can't be recovered, per DeepSeek's FAQ.

Is the DeepSeek API free? No, it bills per million tokens

The DeepSeek API is not free, and DeepSeek's FAQ rules out a cheaper tier: "There is a unified pricing standard and no tiered plans." You top up a balance by PayPal, bank card, Alipay or WeChat Pay. Each call then takes tokens times price from it.

Per 1M tokens, DeepSeek APIV4.1 Flash off-peakV4.1 Flash peakV4 Pro off-peakV4 Pro peak
Input, cache hit$0.003$0.006$0.022$0.044
Input, cache miss$0.15$0.30$0.66$1.32
Output$0.60$1.20$1.98$3.96

Those are the prices on DeepSeek's pricing page as published on 24 September 2026. The $0.15 floor is for fresh input. Input that DeepSeek has already cached from an earlier call, such as a repeated system prompt, costs $0.003. Off-peak is half of peak. Peak is 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, and Chinese public holidays are off-peak all day, so in a normal week the half price applies for 133 hours, or 79% of the time. DeepSeek cut prices to these levels on 10 September 2026, when V4.1 Flash launched, and the page warns that "Product prices may vary".

Two lines in DeepSeek's FAQ make a prepaid balance less of a gamble. Top-up balances don't expire. Unused balances can be refunded. The pricing page also mentions a "granted balance" that's spent before your own money, but DeepSeek publishes no amount for it, so don't count on free credits.

Limits are generous for one person. Each account can run 2,500 requests at once on V4.1 Flash and 500 on V4 Pro, per DeepSeek's rate limit page, and every model has a 1M token context window.

Is DeepSeek free to run locally?

The current model is free to run locally in licence but rarely in practice. DeepSeek publishes V4.1 Flash on Hugging Face under the MIT license, which allows commercial use. The weights are 48 files totalling 510 GB, for a model DeepSeek describes as 552B parameters. That's server hardware, not a laptop.

Ollama, the usual way to run models on a personal computer, lists V4.1 Flash only with a cloud tag on 12 October 2026. A cloud tag runs on Ollama's servers, not yours.

What fits on a laptop is the older R1 family. Ollama's deepseek-r1 page lists distilled versions from 1.1 GB (1.5B) to 43 GB (70B), and the default 8B tag is 5.2 GB with a 128K context window. They run offline at no cost, and nothing you type leaves the machine. Getting one going takes three steps:

  1. Install Ollama from ollama.com.
  2. Run ollama run deepseek-r1:8b in a terminal. The first run downloads 5.2 GB.
  3. Type your question at the prompt. Close the terminal to stop it.

These distills are a much smaller model from an earlier generation than the one the app serves. They suit privacy and offline use. For open models more broadly, the open source LLM guide compares the families.

Paying for DeepSeek somewhere else rarely beats free

For DeepSeek alone, the free app beats every paid route, and the API beats every subscription. 2,500 V4.1 Flash chat turns cost about $1.13 off-peak on DeepSeek's API. Whizi's Pro plan is $29.99 a month for 2,500 credits, and its five DeepSeek models cost 1 or 2 credits a message, so on DeepSeek price alone Whizi loses by a wide margin. What Pro buys is DeepSeek next to Claude Opus, Claude Sonnet and ChatGPT Sol in one chat, served through OpenRouter and routed providers instead of DeepSeek's own app. Starter and the free Whizi account have no DeepSeek model at all. DeepSeek in Whizi lists each model and its credit cost.

Workflow checklist
  • Use chat.deepseek.com or the official app, which list no price and no in-app purchases
  • Turn off Search before retrying when the busy message names the search service
  • Use the training opt out in DeepSeek's privacy policy before the free chat sees anything personal
  • Schedule API jobs outside 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays to pay half
  • Run the 5.2 GB R1 8B distill through Ollama when nothing may leave your machine
Common questions

Frequently asked questions

Is DeepSeek free on OpenRouter?

No, not as of 12 October 2026. OpenRouter's model list carried 18 DeepSeek model ids that day and none was a :free variant. V4.1 Flash listed at $0.30 in and $1.20 out per million tokens, which matches DeepSeek's own peak price. Free variants on OpenRouter come and go, so check the model page before you build on one. If you use OpenRouter without building an app, see the best OpenRouter alternative.

Does the DeepSeek API give new accounts free credits?

DeepSeek doesn't publish one. Its pricing page says charges come from a granted balance first when you have one, but it names no amount and no rule for getting it, and its FAQ says there are no tiered plans. Plan on topping up. Balances don't expire, and unused money can be refunded. DeepSeek payment methods lists how to top up the balance.

Is DeepSeek Expert Mode free?

Yes. Expert Mode on the website and app runs DeepSeek V4 Pro, per DeepSeek's 13 August 2026 release note, and the app lists no in-app purchases. Through the API the same V4 Pro costs $0.66 to $1.32 per million input tokens and $1.98 to $3.96 per million output tokens, as of September 2026. DeepSeek vs Gemini compares the two on score and cost per task.

When is the DeepSeek API cheapest?

Off peak. DeepSeek doubles its prices from 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, except on Chinese public holidays. Outside those hours its Flash model costs $0.15 per million fresh input tokens and $0.60 per million output tokens, as of October 2026. DeepSeek vs Grok compares the cost per task.