Skip to content

DeepSeek AI: V4 Models, Pricing, Free Chat and Privacy

A woman using a laptop with a simple AI chat interface and a small blue whale icon on the screen in a bright home office


A woman using a laptop with a simple AI chat interface and a small blue whale icon on the screen in a bright home office

DeepSeek AI is a free AI chatbot and low-cost API from a Chinese lab in Hangzhou, built on open-weight models that now run under the V4 name.

You can use it three ways: free chat in a browser or phone app, a pay-as-you-go API, and downloaded model weights you run on your own hardware. Liang Wenfeng founded the lab in 2023, and it grew out of High-Flyer, a quantitative trading firm. The official chat runs at chat.deepseek.com, with phone apps alongside it.

What DeepSeek AI is

DeepSeek AI is both a company and a family of large language models. The lab publishes its model weights under the MIT license, so anyone can download, study, and run them, including for commercial work. Developers could test the full model instead of trusting a demo, which is why it spread so fast.

The current generation is the V4 family. Both API models read up to 1 million tokens of context in one request and can return up to 384,000 tokens of output. Each runs in fast and thinking modes.

It works like ChatGPT or Claude on the surface: you type a prompt and get a reply. The difference is underneath: open weights and price. The ChatGPT alternative guide lists other free chatbots, and the Claude vs ChatGPT comparison covers the closed models most people compare DeepSeek against.


A close-up of a laptop screen showing blank chat bubbles and two toggle switches on a dark blue chat interface

How to use DeepSeek for free

Open chat.deepseek.com and sign up with an email address or a Google or Apple account. The same account works in the phone apps. Chat is free, with no paid consumer tier.

Two toggles sit near the message box. DeepThink switches on step-by-step reasoning, which suits math, logic, and code the fast mode got wrong, at the cost of slower replies. Search lets the model look up current information, which helps with news and recent prices. Leave both off for quick drafting. Labels change from time to time, so trust the buttons on screen.

You can also attach files to a chat and ask questions about them. The free AI tools for students roundup lists no-cost options for other tasks.


An abstract diagram of glowing connected nodes and branching paths flowing into one bright answer on a dark blue background

DeepSeek V4-Flash vs V4-Pro: which model to pick

The API lists two models. deepseek-flash runs DeepSeek-V4.1-Flash-1017, the fast model. deepseek-v4-pro runs DeepSeek-V4-Pro-0813, the larger model. The older name deepseek-v4-flash still works, but that model is retired, and requests under the old name now run on V4.1-Flash at the Flash price. A guide that still leads with V3 or R1 describes an older generation.

Pick Flash for everyday work: chat, drafting, extraction, and light coding. It is cheaper and the only one of the two that reads images. Pick Pro for hard reasoning and complex coding, where a wrong answer costs more. Pro costs about 3.3 times as much per output token and 4.4 times as much per input token.

The odd part is that Pro is not better at everything. Flash reads images and Pro does not, so a task that starts with a screenshot has only one choice. Start with Flash, and move a task to Pro only when Flash gives wrong or shallow answers. Other labs split their models into fast and strong tiers too, as the Gemini AI Studio guide shows.


A developer working at two monitors showing colorful blurred lines of code in an evening office

DeepSeek API pricing and peak hours

The API is prepaid and billed per million tokens, with no monthly fee. You create an account on the DeepSeek platform, add a balance, and create an API key. It accepts the same code shape as the OpenAI API, so existing tools need only a new base URL and model name.

Flash costs $0.15 per million input tokens off-peak and $0.30 at peak, with output at $0.60 and $1.20. Pro costs $0.66 off-peak and $1.32 at peak for input, with output at $1.98 and $3.96. Cached input costs far less: $0.003 off-peak for Flash and $0.022 for Pro. Peak hours run 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. Every other hour bills at the off-peak rate, half the peak rate.

Prices change often, so check before you budget a large job. DeepSeek lists the current model prices and peak hours on its pricing page.


A top-down view of a developer desk with a laptop showing a blurred terminal window, a notebook, a phone, and a cup of coffee

Privacy: what happens to your data

Read this before you paste anything sensitive. DeepSeek is a Chinese company, and its privacy policy says it collects, processes, and stores personal data in the People’s Republic of China. What you type and upload counts as personal data, along with device details and chat history. Inputs are kept for as long as you have an account, and they may be used to improve its models.

Do not paste passwords or API keys, customer records, medical or financial documents, or code your employer owns. Treat the free chat like a public notebook: fine for drafts, wrong for secrets. Geodd covers the same privacy points in its DeepSeek guide. Several countries restrict it on government devices.

The private path is to run the model yourself: a copy on your own computer sends nothing to DeepSeek.


A glowing blue shield with a padlock floating in a dark server room between rows of server racks

Running DeepSeek on your own computer

Ollama is the simplest route. Install it, pull a DeepSeek model with one command, and chat in your terminal or a desktop app. Smaller distilled versions, from 1.5 billion to 70 billion parameters, run on a laptop or one strong graphics card. The full flagship models need a multi-GPU server, so start small. Ollama is free to download for Mac, Windows, and Linux.

A local model keeps every prompt on your machine and costs nothing per message after the hardware. The trade is ability: a small distilled model answers less well than the hosted V4 models. Test it on your real tasks before you rely on it.

DeepSeek earns its place as a cheap, open alternative to the big closed chatbots, as long as you respect the privacy line and pick the right model. The best AI tools compared puts it in a wider field, and the Meta AI app guide covers another free chatbot.


A laptop showing blank chat bubbles beside a small home server box with a blue light on a wooden desk

Frequently asked questions about DeepSeek

What is DeepSeek AI?

DeepSeek AI is a free chatbot and low-cost API from a lab in Hangzhou, built on the open-weight V4 model family.

Is DeepSeek AI free?

Web and phone chat is free, with no paid consumer tier. The API is prepaid per token, and self-hosting costs nothing per message beyond your hardware.

Which DeepSeek model should I use, Flash or Pro?

Use Flash for everyday work and images. Use Pro for hard reasoning and complex coding. Start with Flash and switch only when it fails.

Does DeepSeek store data in China?

Yes. Its privacy policy says personal data is collected, processed, and stored in the People’s Republic of China. Do not paste credentials, customer records, or proprietary code into it.

Can I run DeepSeek on my own computer?

Yes. Install Ollama and pull a distilled model, such as an 8 billion parameter version that runs on a modern laptop. Your prompts stay on your machine.

What is DeepThink in DeepSeek?

DeepThink is the reasoning toggle in DeepSeek chat. It shows step-by-step thinking before the final answer. It is slower, so use it for math, logic, and hard coding problems.

Leave a Comment