DeepSeek for free: chat, API limits and service status

DeepSeek is free to use in its web chat and mobile app: the official download page says you can chat with the assistant for free. The API is not free, and its tokens are deducted from a balance you top up. The model API limits concern concurrent requests, not the number of 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 unavailable.

What is free at DeepSeek: chat, app, Harness, weights

A summary from the official DeepSeek pages, as of 10 November 2026:

The "Deduction Rules" section mentions a "granted balance", meaning a balance issued by DeepSeek, which is spent first. It is not a permanent free tier, and the docs do not say what these sums are or how to get them. Top-up balances do not expire, according to the DeepSeek FAQ. One detail changed since our previous check on 29 October: search and fetch through Harness were free until 00:00 Beijing time on 1 November 2026, and the price list now bills them per request.

The web chat has two quirks. The download page calls it free but publishes no limits on the number of messages or the length of a conversation. Third-party blogs quote various numbers, and we would not rely on any of them. And the chat lives apart from the API: they have different rules and different ways to pay.

DeepSeek limits: what is published and what is not

For the chat and the app, the official pages we opened state no limits. What the interface itself shows you matters more than any figure from outside.

For the API, limits are published on the Rate Limit & Isolation page:

What to know about these rules:

• A request holds a slot from the moment it is sent until the model finishes its reply. Beyond the limit you get HTTP 429.
• The limit is counted per account, not per key. Several keys do not raise it.
• You can ask for more slots, and DeepSeek says capacity expansion costs nothing extra.
• The user_id parameter separates the users of your service: it isolates their cache and scheduling, and on accounts with an increased limit it adds a per-user_id limit 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.
• Search and Fetch are counted separately from the model limits: Search is capped at 100 QPS per account, Fetch at 25 QPS per account plus a per-domain QPS limit, and DeepSeek does not publish the per-domain number. Above the cap you get HTTP 429.

Cost also depends on the time of day. Off-peak prices are half of peak ones. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. Weekends and Chinese holidays are off-peak in full.

DeepSeek Platform: key, balance and errors

Keys are issued by the API platform (platform.deepseek.com), and the docs give platform.deepseek.com/api_keys as the key page. The request address is https://api.deepseek.com for the OpenAI format and https://api.deepseek.com/anthropic for the Anthropic format. The model name is deepseek-flash or deepseek-v4-pro.

Prices per 1M tokens off-peak: deepseek-flash $0.15 for input on a cache miss and $0.6 for output, deepseek-v4-pro $0.66 and $1.98. At peak it is exactly double. The DeepSeek FAQ lists PayPal, bank card, Alipay and WeChat Pay for top-ups, and how that works from your country is covered in DeepSeek API: key, payment and prices from restricted regions.

Three errors turn up most often, and each points to its own cause:

The same error page documents 400 and 422 (a malformed body and invalid parameters), 500 (a server error, retry after a short wait) and 503 (a server overloaded by high traffic, retry after a short wait).

DeepSeek status: how to tell whether the service is down

The official status page is status.deepseek.com. It is linked from the API docs header (the API Status item) and from the main site. The page shows separate indicators:

• DeepSeek V4 Pro API;
• DeepSeek V4.1 Flash API;
• the chat (Chat Service);
• file upload;
• search, along with the Search API and Fetch API.

You can subscribe to updates, and the page also offers feeds at status.deepseek.com/feed.rss and status.deepseek.com/feed.atom. If you depend on the API, subscribe to a feed so you do not find out from your logs.

When DeepSeek "does not respond", go in this order:

1. Open status.deepseek.com and read the indicator for the service you use. The chat and the API are separate there.
2. Compare the error code. 401, 402 and 429 point to your settings, your balance and your request rate, not to an outage.
3. Send one minimal request from another machine or another route. If it goes through, your client or your network is at fault.
4. If the status is green but requests time out, do not wait for the page. On 14 September 2026 users reported an outage while the page stayed green. The full order of checks, the causes by symptom and the list of recent incidents are in DeepSeek not working: how to check status and fix it.

Download and install DeepSeek: app, Harness, local models

"Download" can mean three different things. All three start from the same place, the download page on deepseek.com.

Search results for "download deepseek" can include clones. Take files only from the deepseek.com page, and treat an APK from any other source as someone else's build.

Harness also runs without the desktop installer. Its GitHub README, which describes a developer preview with compatibility-breaking changes expected, gives npx @deepseek-ai/dsh web for Node.js users: it starts the web UI at http://127.0.0.1:3080.

Running locally does not replace the chat. The weights of the current models are very large: the V4.1-Flash model card cites 552B backbone parameters, and V4 Pro has 1.6T. For an ordinary computer people install the previous generation: the Ollama library has the deepseek-r1 family in sizes 1.5b, 7b, 8b, 14b, 32b, 70b and 671b, with a command such as ollama run deepseek-r1:8b.

DeepSeek in your own language, free

The language switch on deepseek.com offers Chinese and English. We found no Russian or other interface on the site. You can write prompts in other languages, and you should judge answer quality on your own tasks: we have no single measurement across all versions.

If you need an interface in your language around the model, connect it to a client that has your localization: the model connects by key.

The free DeepSeek model in Claudexia

The Claudexia catalog has deepseek-v4.1-flash-1017-free. In the public catalog on 10 November 2026 its price is $0 for input, output and cache, with a context of 1,000,000 tokens. The terms and limits of the free model are on the models page: the catalog gives no numeric limits, and we do not invent any. General gateway rules are in the section on rate limits.

A minimal call:

curl https://api.claudexia.tech/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 the key by following the quickstart. For what else is free around the Claude API and how to pay the minimum, see free Claude API: what exists in 2026. Paid models are paid for by card, by SBP bank transfer in Russia or with crypto.

We would not rely on a free model for production load. The paid deepseek-v4.1-flash-1017 costs $0.08 per 1M input tokens and $0.28 per 1M output tokens in the Claudexia catalog, and deepseek-v4-pro costs $0.06 and $0.18. Those are Premium channel prices on 10 November 2026, and current ones are on the models page. For help choosing between DeepSeek and other families, see Claude vs GPT vs Gemini vs DeepSeek.

FAQ

Is DeepSeek free?

The chat and the app are free, as the official download page says. The API is paid: tokens are deducted from a prepaid balance, and the price list shows 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 the app, no numeric limits are published in the official sources we opened. For the model API the main limit is concurrent requests per account, 2500 for deepseek-flash and 500 for deepseek-v4-pro, and DeepSeek grants an increase on request at no charge. Search and Fetch have their own QPS limits, 100 and 25 per account.

How do I check DeepSeek status?

Open status.deepseek.com. It has separate indicators for the V4 Pro API, the V4.1 Flash API, the chat, file upload, search, the Search API and the Fetch API, plus a subscription and RSS and Atom feeds. If the status is green but requests fail, look at the error code: 401, 402 and 429 are caused by your settings, your balance and your request rate.

Where do I download DeepSeek and how do I install it?

The iOS and Android apps and the Harness desktop 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 for your own computer are installed through Ollama or Hugging Face, but the current DeepSeek models are too large for an ordinary machine.

What is DeepSeek Platform?

The API platform at platform.deepseek.com, where you create keys and top up the balance. The chat at chat.deepseek.com is a separate thing. The API needs a key and a prepaid balance, and the request address is https://api.deepseek.com.

Can I use the DeepSeek API for free?

There is no permanent free tier in the price list. The only thing mentioned is the "granted balance", a balance issued by DeepSeek that is spent first. A free option for API requests comes from the Claudexia catalog: the model deepseek-v4.1-flash-1017-free, with its terms on the model page.

Does DeepSeek work in my country?

We found no official list of countries with restrictions. The payment side, using Russia as the example, is covered in DeepSeek API: key, payment and prices from restricted regions. If paying DeepSeek directly is not convenient, the models can be taken through a gateway with local payment: a Claudexia key, the address https://api.claudexia.tech/v1 and an ID from the catalog.