Updated 2026-11-20
How to Use DeepSeek V4: Chat, API or Local Deployment
For a single question, start in DeepSeek’s web chat. To process a queue of questions from your own application, use the hosted API. If the input must stay on hardware you control, first check whether your machine can run the specific checkpoint you need.
| Access option | What you need | Where to start |
|---|---|---|
| Official chat | A browser and a DeepSeek chat sign-in | Open chat.deepseek.com |
| Official API | A DeepSeek platform account with credit and an API key | Create a key on platform.deepseek.com and make a first request |
| Local deployment | The weights, a compatible inference runtime and enough accelerator memory | Read the checkpoint’s model card and check your hardware before downloading |
Use official chat for an individual question
Open DeepSeek’s chat site, start a conversation and enter your question. You can ask for a revision in the same thread. This is the shortest path for someone working in a browser and requires no API key or local model download.
Use the official API for a script or application
Create a key in DeepSeek’s API platform, store it on the server and use https://api.deepseek.com as the base URL for an OpenAI-compatible client. Set model to deepseek-flash for text or image input, or deepseek-v4-pro for text. Send one short request and check the returned answer and token usage before connecting a batch job.
DeepSeek bills direct API use to your platform account by input and output tokens. If you use a Coding Plan sold by this independent hub, use the key and base URL supplied with that plan and check its allowance and billing terms on the pricing page.
Use local weights when you control the machine
Choose the exact checkpoint and read its model card before downloading the weights. Check that your runtime supports that architecture and your accelerator, then budget storage and memory for the weight format and the context length you plan to use. A tutorial for the original V4 Flash can target a different architecture from Flash-1017.
Record the weights revision and runtime version, then try a short prompt with the networking settings you intend to use. For a completely local workflow, check where the client, inference server and any external tools send the input.
FAQ
Do I need to buy a Coding Plan to use DeepSeek chat or its direct API?
You can open official chat directly or set up your own API account with DeepSeek. This hub’s Coding Plans are a separate way to purchase configured access.
Does downloading Flash include the hosted chat features?
The download supplies the model weights. If you need file parsing, web search or a chat interface in your deployment, you also need software that provides those features.
If you already know how you will connect, check which model name your client or runtime expects before choosing a checkpoint.