You can use DeepSeek for free in the web chat and in the mobile app: the official download page says you can talk to the assistant at no charge. The API works differently, because tokens are deducted from a balance. API limits count concurrent requests, not messages. Below: what is free, how to check DeepSeek status, how to install the app without picking up a fake, and what to use when the chat is down.
What is free in DeepSeek: chat, app, Harness and weights
A summary as of 29 October 2026, taken from the official DeepSeek pages:
| What | Free? | Where it says so |
|---|---|---|
| Web chat at chat.deepseek.com | Yes | Download page: "Chat with leading AI models for free" |
| iOS and Android app | Yes, same wording | Same page |
| DeepSeek Harness (agent, desktop and npx) | The tool is free, tokens are paid | Project README (MIT), pricing table |
| API on the DeepSeek platform | No, prepaid | Pricing table, "Deduction Rules" section |
| Open weights on Hugging Face | Yes, MIT license, but you need hardware | Model cards |
| Search and page fetch through Harness | Free until 00:00 Beijing time on 1 November 2026 | Pricing table |
The "Deduction Rules" section mentions a "granted balance", meaning a balance DeepSeek hands out itself. When charges are deducted, it is spent first. That is not a permanent free tier, and the documentation does not say where such amounts come from or how to get one.
Two things to know about the web chat. The download page calls it free but publishes no limit on messages or conversation length. Third-party blogs quote various numbers, and we would trust none of them. Also, the chat lives apart from the API, with its own rules and its own payment methods.
DeepSeek limits: what is published and what is not
In the official pages we opened, there are no limits for the chat or the app. Go by what the interface itself shows, not by someone else's figure.
For the API, limits are described on the Rate Limit & Isolation page:
| Parameter | deepseek-flash | deepseek-v4-pro |
|---|---|---|
| Concurrent requests per account | 2500 | 500 |
| Context | 1M tokens | 1M tokens |
| Maximum response | 384K | 384K |
What follows from it:
- A slot is held from the moment a request is sent until the model finishes its answer. Anything over the limit gets HTTP 429.
- The limit applies to the account, not the key, so extra keys add nothing.
- You can ask for more slots. DeepSeek says the extension costs no extra money.
- The
user_idparameter separates your service's users: it isolates their cache and scheduling, and on accounts with an extended limit it adds a per-user_idlimit as well. - While a request waits in the queue, the server holds the connection and sends empty lines or SSE comments
: keep-alive. If inference has not started within 10 minutes, the connection is closed.
Price also depends on the hour. Off-peak it is half: peak is 01:00-04:00 and 06:00-10:00 UTC on weekdays, which is 04:00-07:00 and 09:00-13:00 Moscow time. Weekends and Chinese public holidays run entirely at the lower price.
DeepSeek Platform: key, balance and errors
Keys come from the API platform (platform.deepseek.com); the documentation names the key page as platform.deepseek.com/api_keys. Requests go to https://api.deepseek.com in the OpenAI format and to https://api.deepseek.com/anthropic in the Anthropic format. The model is given as deepseek-flash or deepseek-v4-pro.
Prices per 1M tokens off-peak: deepseek-flash $0.15 for uncached input and $0.6 for output, deepseek-v4-pro $0.66 and $1.98. At peak it is exactly double. The models themselves are covered in detail in the DeepSeek 4.1 Flash review.
Three errors come up most often, and each points to its own cause:
| Code | Cause per the documentation | What to do |
|---|---|---|
| 401 | Invalid API key | Check the key or create a new one |
| 402 | Balance exhausted | Top up the account |
| 429 | Requests too frequent or concurrency limit exceeded | Slow down, add a queue |
How to top up the balance from Russia is covered in DeepSeek API from Russia.
DeepSeek status: how to tell the service is down
The official status page is status.deepseek.com. It is linked in the API documentation header (the API Status item) and on the site's home page. The indicators there are separate:
- DeepSeek V4 Pro API;
- DeepSeek V4.1 Flash API;
- chat (Chat Service);
- file upload;
- search, plus the Search API and Fetch API.
You can subscribe to updates, and RSS and Atom feeds exist. If your product depends on the API, subscribe to the feed: the notice will reach you before you spot the failure in your logs.
When DeepSeek "doesn't respond", work through this:
- Open status.deepseek.com and find the indicator for the service you use. Chat and API are separate there.
- Look at the error code. 401, 402 and 429 point at your settings, balance and request rate, not at an outage.
- Send one minimal request from another machine or over another route. If it goes through, your client or network is the culprit.
- Status is green but requests time out? Don't wait for the page to update. The case of 14 September 2026, when users reported an outage while the page stayed green, is covered in DeepSeek V4 is not working.
Download and install DeepSeek: app, Harness, local models
"Download" can mean three different things. They share one source: the download page on deepseek.com.
| What | Where from | Conditions |
|---|---|---|
| iOS app | The link on the download page leads to the App Store (cn storefront) | We did not check whether it opens with your Apple ID |
| Android app | The page links to an APK on the download.deepseek.com domain | Installed by hand, so check the domain |
| DeepSeek Harness (agent) | Same page, Harness section | macOS 13+ on Apple silicon, or 64-bit Windows |
| Model on your own computer | Hugging Face, Ollama | Tens of gigabytes and a GPU, depending on the model |
A search for "deepseek download" also returns clones. Take files only from the deepseek.com page, and treat an APK from anywhere else as someone else's build.
Running locally is no substitute for the chat. The weights of the latest models are too large: V4.1-Flash has 552 billion parameters and V4 Pro has 1.6 trillion. An ordinary computer gets the previous generation. The Ollama library has the deepseek-r1 family in sizes 1.5b, 8b, 32b, 70b, 671b and others, and the command looks like this: ollama run deepseek-r1:8b. We cover the Harness agent in a separate article, DeepSeek Harness.
DeepSeek in Russian, free
On the deepseek.com home page, the language switcher offers Chinese and English. We did not find a Russian interface on the site. You can write prompts in Russian, but judge the answer quality on your own tasks: we have no single benchmark across all versions.
Need a Russian-language interface around the model? Connect it to a client with Russian localization. The DeepSeek Harness web interface ships in Russian among other languages, and the model plugs in by key.
The free DeepSeek model in codexkey
The codexkey catalog has deepseek-v4.1-flash-1017-free. In the 29 October 2026 price list it costs $0 for input, output and cache, with a 1000K context. For the terms and limits of the free model, see the models page: the catalog gives no limit figures, and we won't invent any. The gateway's general rules are in the section on rate limits.
A minimal call:
curl https://api.codexkey.online/v1/chat/completions \
-H "Authorization: Bearer sk_cdx_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "deepseek-v4.1-flash-1017-free", "messages": [{"role": "user", "content": "Answer in one sentence: what is a cache hit in an API?"}]}'
You create a key by following the quickstart. Paid models are paid for with a Russian bank card, SBP (the Russian Faster Payments System) or crypto. The catalog holds two more free models, qwen3.8-flash-free and minimax-m3-free. How to run a coding agent on such models is shown in Claude Code for free.
We would not trust the free model with a production workload. The paid deepseek-v4.1-flash-1017 costs $0.04 per 1M input tokens and $0.16 per 1M output tokens in codexkey, and deepseek-v4-pro costs $0.12 and $0.34. That is the codexkey price list for 29 October 2026; current prices are on the models page.
FAQ
Is DeepSeek free?
The chat and app are free, as the official download page states. The API is paid: tokens are deducted from a prepaid balance, and the price list has no permanent free tier. DeepSeek Harness is free as a program, but the model inside it is paid for.
Does free DeepSeek have limits?
For the chat and app, no numeric limits are published in the official sources we opened. The API has one limit: concurrent requests per account, 2500 for deepseek-flash and 500 for deepseek-v4-pro. DeepSeek grants an extension on request, and for free.
How do I check DeepSeek status?
Open status.deepseek.com. It has separate indicators for the V4 Pro API, the V4.1 Flash API, chat, file upload, search, the Search API and the Fetch API, along with update subscriptions, RSS and Atom. Status is green but requests fail? Look at the error code: 401, 402 and 429 come from settings, balance and request rate.
Where can I download DeepSeek and how do I install it?
The iOS and Android apps and the desktop Harness agent for macOS and Windows are listed on the deepseek.com/download page. Treat other sites and APKs from third-party sources as someone else's builds. Models can be installed on your own machine through Ollama or Hugging Face, but the latest DeepSeek models are too large for an ordinary computer.
What is DeepSeek Platform?
It is the API platform at platform.deepseek.com, where you create keys and top up the balance. The chat at chat.deepseek.com is set up separately. The API needs a key and a prepaid balance, and requests go to https://api.deepseek.com.
Can I use the DeepSeek API for free?
The pricing table has no permanent free tier. The only thing it mentions is the "granted balance", a balance issued by DeepSeek that is spent first. A free option for API requests exists in the codexkey catalog: the model deepseek-v4.1-flash-1017-free, with terms on the model page.
Does DeepSeek work in Russia?
We found no official list of restricted countries. Paying for the API from Russia is covered in DeepSeek API from Russia. If paying DeepSeek directly is awkward, you can take the models through a gateway that accepts rubles: a codexkey key, the address https://api.codexkey.online/v1 and an ID from the catalog.