DeepSeek AI: Chat, App, API, and Models — An Independent Guide
DeepSeek is an AI company and product ecosystem built around generative AI models, an official chat assistant, mobile apps, developer APIs, and open model resources. This page explains each of those parts, what it costs, and which one fits what you are trying to do — with direct links to DeepSeek’s own services.
Search results tend to mix the chat, the app, the API, the models, and third-party tools built on DeepSeek models. They are priced, supported, and limited differently, so the first useful step is telling them apart. This page is not the official DeepSeek website and does not host an on-site chat.
Quick answer. DeepSeek’s consumer chat and app are free to use. The API is billed per token: DeepSeek V4.1 Flash starts at $0.15 per million input tokens and $0.60 per million output tokens off-peak, with peak hours billed at double those rates. The models are also published as open weights, which you can run yourself if you have the hardware. Most readers need one of those three routes, not all of them.
Last updated 21 November 2026 · Source documentation checked 31 October 2026 · API prices and request fees rechecked 18 November 2026
What is DeepSeek?
DeepSeek can refer to several related things: the company, the official AI assistant, the mobile app, the API platform, the underlying models, and the open model resources published under the DeepSeek name. They share a name and a model family, but they are accessed and priced differently — and a reader looking for one of them can easily land on a page about another.
| DeepSeek term | What it means | What it costs | Where to go |
|---|---|---|---|
| DeepSeek chat | The official web assistant for questions, writing, coding help, and general AI tasks. | Free to use. | chat.deepseek.com |
| DeepSeek app | The official mobile assistant for iOS and Android. | Free to use. | Official download page |
| DeepSeek API | Developer access to hosted DeepSeek models. | Billed per token; Flash input starts at $0.15 per million tokens off-peak. | Platform and API docs |
| DeepSeek models | The models behind the chat and the API, at different capability and price tiers. | Included in the hosted products; the Pro tier costs more per token than Flash. | Model guides and API docs |
| Open model resources | Model cards, code, and repositories published by DeepSeek. | Free to download; you pay for GPUs, hosting, and operations. | GitHub and Hugging Face |
DeepSeek’s official website currently highlights DeepSeek-V4.1-Flash and links to the web chat, API platform, API docs, downloads, DeepSeek Harness, research pages, and service status. Use the official site when you need the most direct route to a product; use this guide when you need to work out which product you actually want.
How much does DeepSeek cost?
Pricing depends on how you reach DeepSeek, not on how much you use the name. The consumer chat and app are advertised as free. The API is billed per million tokens, split into input that misses the cache, input that hits the cache, and output. The rate also depends on the model and on the hour you send the request.
The table below is arithmetic, not a quote. It applies DeepSeek’s published Flash and V4 Pro rates to a request shape of 2,000 input tokens and 500 output tokens, which is the same shape as the worked example on our DeepSeek pricing page. Change the request shape and every number moves.
| Requests per month | Flash, off-peak | Flash, off-peak, 70% cache hits | Flash, peak | V4 Pro, off-peak |
|---|---|---|---|---|
| 10,000 | $6.00 | $3.94 | $12.00 | $23.10 |
| 100,000 | $60.00 | $39.42 | $120.00 | $231.00 |
| 1,000,000 | $600.00 | $394.20 | $1,200.00 | $2,310.00 |
Computed from the official rates effective 10 September 2026 for Flash and 16 August 2026 for V4 Pro. Excludes Search and Fetch request fees, retries, failed generations, taxes, and any infrastructure you run around the API.
Four things move that bill, and only one of them is the model you choose:
- Model tier. V4 Pro is billed at several times the Flash rate for identical token counts. Extraction, classification, and most summarisation work does not need it.
- Cache-hit rate. Context you resend on every call can be billed at the cached-input rate instead of the full input rate. If your prompts share a long system message or document, this is usually the largest single saving.
- Timing. Off-peak rates are half the peak rates. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday. A nightly batch job can cost half as much without producing different output.
- Output length. Output is the most expensive line on most bills. Fixed answer lengths and clear stop conditions cut more than trimming prompts usually does.
The pricing page lists every rate by model and billing period, and works through an example you can copy for your own request volume.
Which DeepSeek option fits your budget and scale?
Most readers arrive in one of five situations. The right route is different for each, and the gap between them is wide enough to choose deliberately rather than by habit.
| Your situation | What you actually need | Realistic monthly cost | Next step |
|---|---|---|---|
| You only want to ask questions | The free consumer chat or app | $0 | Official chat |
| You are prototyping in code | Pay-as-you-go API at low volume, one API key, a spend cap | About $6 at 10,000 calls a month on Flash at off-peak rates | API key guide |
| You are shipping to real users | API plus context caching, per-key budgets, logging, and a fallback model | Scales with tokens; model and cache decisions matter more than the provider | Pricing and caching |
| You cannot send data to a hosted API | Open weights on your own or rented GPUs, plus the operations work | A fixed monthly GPU bill, with no per-token charge | Run DeepSeek locally |
| You are evaluating it for a team | The same API, plus a review of processing location, retention, limits, and failure behaviour | Depends on volume; budget the review time as well as the tokens | Business checklist |
The last three rows are where the budget actually lives. A prototype that costs a few dollars a month can become a production line item quickly, and the two decisions that move it most are the model tier you send work to and how much input you can serve from cache rather than resend.
Where to go next, by task
- I want to chat with DeepSeek: the official web chat is the fastest route for questions, writing, and coding help.
- I want the mobile app: use the official download page rather than a third-party mirror.
- I want to build with the API: start with the API key guide, then read the API documentation guide.
- I want to run it myself: the local setup guide covers hardware, quantisation, and what changes when you stop paying per token.
- I want to compare tools: the comparison hub covers ChatGPT, Claude, Gemini, Grok, Kimi, and Qwen.
The official DeepSeek chat and app
The official DeepSeek web chat is available at chat.deepseek.com. The official download page describes the DeepSeek app as the official AI assistant from DeepSeek and links to the iOS and Android options. It currently says the app lets users chat with DeepSeek models for free.
That free claim covers the consumer app, not the API. Developer API calls are billed according to the API platform’s current rates, and a free app account and a funded API account are different things. If you are budgeting, read the API pricing rather than the app description.
When you search for DeepSeek downloads, use the official page rather than a third-party mirror. That reduces the chance of installing an outdated, modified, or unofficial application, and it avoids handing account credentials to a site that is not DeepSeek.
DeepSeek models and the API for developers
Developers reach DeepSeek through the official API platform. The current documentation says the API accepts OpenAI-compatible and Anthropic-compatible request formats depending on the base URL and configuration, and the quickstart lists deepseek-flash and deepseek-v4-pro as the current model names.
- OpenAI-compatible base URL:
https://api.deepseek.com - Anthropic-compatible base URL:
https://api.deepseek.com/anthropic - API keys: created and managed through the DeepSeek Platform.
- Documentation: use the official API docs for endpoints, models, pricing, rate limits, errors, vision, tool calls, context caching, and agent integrations.
API pricing is usage-based and changes by model, token type, and current rate-card rules. DeepSeek changed both model names and rates several times during 2026 — legacy names such as deepseek-v4-flash are still accepted but route to the current model at the current price. Check the official models and pricing page before budgeting an application, and check it again before a launch.
Running DeepSeek yourself: when local beats the API
The open weights are free to download, but the cost does not disappear — it moves from per-token billing to hardware and operations. Whether that is cheaper depends on how much you use, how much GPU memory the model and your context length need, and whether you count the time spent running it.
Before comparing prices, count the parts that do not appear on an API bill:
- GPU hours, including idle time. Rented capacity is billed whether or not a request arrives.
- Storage, egress, and snapshots for weights, logs, and anything you cache yourself.
- Operations time for upgrades, monitoring, and the day the model does not start.
- Evaluation work to confirm the self-hosted model behaves like the hosted one on your tasks.
- Security work that a provider would otherwise carry: patching, isolation, access control.
The usual way to decide is to compute your monthly API cost for your real request shape, then compare it with a fixed monthly GPU bill. If you stay consistently below the break-even volume, the API is usually cheaper once operations are priced in. The local setup guide covers the hardware side; the cost table above covers the API side.
Can you use DeepSeek for business and teams?
The API is usable for commercial work, and the questions that matter are the same ones you would ask of any hosted model: where the data is processed and stored, how long it is retained, what is logged, what rate limits and failure behaviour you get, and how pricing changes are announced.
- Processing location and retention. The privacy policy says personal data is collected, processed, and stored in the People’s Republic of China. Confirm that this is acceptable for the data you intend to send.
- What you can send. Do not send confidential documents, customer data, secrets, or identity documents unless your own policy allows it.
- Limits and failure behaviour. Know your per-key rate limits and decide what the application does when the API is slow or unavailable. See rate limits.
- Cost controls. Set per-key budgets, log cost per request, and alert before a surprise invoice rather than after it.
- Vendor terms. Ask about pricing-change notice, sub-processors, incident disclosure, and support escalation before you depend on the service.
- Internal approval. Most organisations keep an approved-tool list. Check it before a team connects production data.
None of this is specific to DeepSeek, but the answers are specific to DeepSeek, and they are easier to check before a launch than after one. Our privacy and safety guide covers the policy side in more detail. This section is general information, not legal, security, or compliance advice.
Is DeepSeek free?
DeepSeek’s consumer chat and app experience may offer free access, as described on the official app download page. Developer API usage is a different product: calls are tied to the DeepSeek Platform and its current pricing rules. In practice the answer depends on how you use it:
- Consumer chat or app: check the official chat and app pages for the current access conditions.
- API usage: check the official model pricing page and your account balance in the DeepSeek Platform.
- Open weights: the download is free, but running a model yourself still involves hardware, hosting, engineering, and compliance costs.
- Third-party services: a tool using a DeepSeek model may charge its own prices and apply its own limits, which are not DeepSeek’s prices or limits.
DeepSeek privacy and safety considerations
DeepSeek’s Privacy Policy, last updated 10 February 2026, says the hosted services can collect account information, prompts, uploaded files, photos, feedback, chat history, device and network information, log data, and approximate location based on IP address. It also says the services are provided and controlled by Hangzhou DeepSeek Artificial Intelligence Co., Ltd.
The same policy says personal data is directly collected, processed, and stored in the People’s Republic of China, and that users may have rights depending on where they live, including the ability to opt out of using personal data for training or optimising technologies, subject to applicable rules and exceptions.
For everyday use, a simple rule covers most of it: do not enter confidential work documents, private customer data, secrets, passwords, identity documents, or sensitive personal information unless you are allowed to do so and you understand the applicable terms. For work use, follow your organisation’s approved-tool policy.
DeepSeek’s terms and privacy policy also remind users that generative AI outputs may be inaccurate. Verify important facts, code, numbers, legal claims, medical advice, financial conclusions, and high-stakes decisions before relying on them.
Is DeepSeek the same as ChatGPT?
No. DeepSeek and ChatGPT are separate AI products from different organisations. They overlap in common tasks such as writing, explaining, coding help, summarisation, and question answering, but they use different models, account systems, policies, pricing structures, and feature sets.
If you are deciding between them, compare the specific task rather than the brand name: availability in your region, model behaviour, file and image support, privacy policy, API pricing, coding performance, and your organisation’s approval rules. The comparison hub has a full DeepSeek vs ChatGPT breakdown, plus Claude, Gemini, Grok, Kimi, and Qwen.
Is DeepSeek open source?
DeepSeek publishes model and developer resources through official organisation pages such as DeepSeek on GitHub and DeepSeek on Hugging Face. These resources are useful for researchers, developers, and teams evaluating model releases beyond the hosted chat service.
Use the term open weights carefully. Downloadable weights do not automatically mean that training data, full training infrastructure, all implementation details, or every derivative licence condition is open. Check the licence and model card for the specific release you plan to use — details are in our open weights guide.
Frequently asked questions.
Is Deep-AI.chat the official DeepSeek website?
No. Deep-AI.chat is an independent editorial guide. It is not owned, operated, authorised, or endorsed by DeepSeek, and it does not host an on-site chat, provide accounts, or handle billing. The links to chat.deepseek.com and the download page go to DeepSeek’s own services.
Where can I chat with DeepSeek for free?
At chat.deepseek.com, the official web assistant, or through the official mobile app linked from DeepSeek’s download page. The consumer chat and app are advertised as free to use. Avoid third-party mirrors that ask for your DeepSeek account credentials.
Does DeepSeek have an API, and what does it cost?
Yes. The official API platform is platform.deepseek.com, with documentation at api-docs.deepseek.com. The API is billed per million tokens, separately for cached input, uncached input, and output. Flash starts at $0.15 per million uncached input tokens off-peak, with peak hours billed at double those rates.
How much does DeepSeek cost per month?
There is no monthly API plan; it is pay-as-you-go. A workload of 10,000 calls with 2,000 input and 500 output tokens each costs about $6 a month on Flash at off-peak rates, $3.94 with a 70% cache-hit rate, and $12 at peak rates. Volume scales the bill roughly linearly.
Can I run DeepSeek on my own computer?
Yes. DeepSeek publishes open weights that can be downloaded and run locally, subject to the licence for that specific release. What changes is where the cost sits: instead of paying per token you pay for GPU memory and time, and you take on operations, evaluation, and security work.
Can I use DeepSeek for commercial or business work?
The API is usable for commercial work, but settle four things first: where the data is processed and stored, how long it is retained, what your rate limits and failure behaviour are, and whether your organisation has approved the tool. The privacy policy says personal data is processed and stored in China.
Which DeepSeek model should I start with?
Start with deepseek-flash (DeepSeek V4.1 Flash) for most work: it is the cheaper tier and handles extraction, summarisation, and everyday coding help. Move specific tasks to deepseek-v4-pro only when Flash visibly fails on them, because the Pro tier costs several times more per token.
Can DeepSeek read images?
The current API documentation says deepseek-flash accepts images alongside text for image understanding tasks through the vision capability. Confirm it in the official documentation before relying on it in production, and check whether the interface you are using exposes that capability at all.
Is DeepSeek open source?
DeepSeek publishes model cards, code, and repositories, but the accurate term is open weights. Downloadable weights do not mean that training data, training infrastructure, or every derivative licence condition is open. Check the licence and model card for the release you plan to use.
What should I avoid sharing with DeepSeek?
Avoid confidential work documents, customer data, passwords and API keys, identity documents, medical or financial records, and anything your organisation classifies as restricted. If you need to process that kind of data, use a deployment and a contract that actually cover it.
Official DeepSeek resources
Use these official resources when you need current information or access to DeepSeek services.
| Resource | Official link | Use it for |
|---|---|---|
| DeepSeek website | deepseek.com | Official product and research links. |
| DeepSeek chat | chat.deepseek.com | Official web chat. |
| DeepSeek downloads | Download page | Official app and desktop download routes. |
| API platform | platform.deepseek.com | Developer account, API keys, and platform access. |
| API documentation | api-docs.deepseek.com | Models, endpoints, pricing, errors, guides, and integrations. |
| Pricing | Models and pricing | Current rates by model, token type, and billing period. |
| GitHub | DeepSeek on GitHub | Official repositories and developer resources. |
| Hugging Face | DeepSeek on Hugging Face | Model cards, collections, and model resources. |
| Privacy Policy | Privacy Policy | Data collection, use, rights, and storage information. |
| Terms of Use | Terms of Use | Rules for the hosted DeepSeek services. |
Found an error or an outdated source?
DeepSeek products, model names, prices, limits, and policies change. We update this page when we can verify a change against an official source, and we do not silently rewrite figures. If something here is wrong, or a source has moved, send the URL, the claim, and a reliable source.
