DeepSeek AI

✦ Preview · V4.1-Flash-1017 What's new →

Explore DeepSeek AI.
Explained. Right now.

Everything about DeepSeek AI in one place: the new V4.1 Flash and V4-Pro models, Harness Desktop guides, the free web chat and apps, API pricing, downloads, login help and fixes when DeepSeek isn't working. Independent, and checked against DeepSeek's official sources.

Chat with DeepSeek API Platform Harness Desktop

✦ Getting started

What is DeepSeek, and where should I start?

DeepSeek is a family of open AI models and the tools built around them. Most people meet it through the chat app, then grow into the API or Harness Desktop as their needs get more technical. Here's the lay of the land pick the door that matches what you want to do.

Chat

The simplest way in. Ask questions, draft text, summarise documents or reason through problems in the web or mobile app no setup required. Toggle DeepThink for harder questions and Search for up-to-date answers.

Best for: everyone, first-timers How to use DeepSeek → 02

Models

Under the hood are several models, each with a different job: R1 for step-by-step reasoning, Coder for programming, and V4.1 Flash for fast, low-cost everyday tasks. Knowing which one you're talking to explains most differences in speed, cost and quality.

Best for: choosing the right tool Compare the models → 03

API

Put the same models inside your own apps and scripts. You pay per token, pick the model per request, and can now add web search and fetch so answers are grounded in live pages. Our guide walks through your first call and what it will cost.

Best for: developers & builders Read the API guide → 04

Harness Desktop

An open-source desktop agent that works on your computer, not just in a chat box: it edits code, writes documents and spreadsheets, runs scheduled tasks and extends itself with plugins. It's in public preview and the most capable option on this list.

Best for: hands-on, agentic work Explore Harness →

Not sure? Start with Chat, read the model overview once you notice differences in answers, then move to the API or Harness when you want it working for you automatically.

✦ DeepSeek models

What are the differences between V4 Pro, V4.1 Flash, R1, and Coder?

Four models, four jobs. The table below compares what each one is good at, what you can feed it, and where you can actually use it. Follow the links for benchmarks, limits and pricing on each model's own page.

Model	Best for / use cases	Supported inputs	Where to access
V4 Pro Flagship	The strongest general model. Long, nuanced writing, complex multi-step analysis, large codebases and anything where quality matters more than speed or cost.	Text and long documents; file uploads in the app; large context for whole-project work.	Chat · API · Harness
V4.1 Flash Latest · 1017	Fast, low-cost everyday work: drafting, summarising, Q&A, chat assistants, and high-volume API traffic. The 1017 update sharpened project development and added web search & fetch via the API.	Text and documents; web search & fetch tool calls through the API.	Chat (default) · API · Harness
R1 Reasoning	Hard problems that reward thinking step by step: maths, logic, research questions, planning and tricky debugging. Slower and more verbose, but shows its reasoning.	Text; files via the chat app. Reasoning trace returned alongside the answer.	Chat (DeepThink) · API · open weights
Coder Code	Writing, completing and refactoring code, fill-in-the-middle completions and repository-level tasks. Tuned on code rather than general conversation.	Code and text; repository context; FIM prompts through the API.	API · Harness Coding · open weights / IDE plugins

Rule of thumb

Start with V4.1 Flash. Switch to R1 when the answer needs careful reasoning, to Coder when the task is pure code, and to V4 Pro when you'd pay more for the best possible result.

Head-to-heads

V4 Pro vs V4.1 Flash →
DeepSeek vs ChatGPT →
API pricing by model →

✦ DeepSeek Harness guide

What is Harness, and how do I use it?

Harness Desktop is DeepSeek's open-source desktop agent. Instead of answering in a chat box, it works inside your files and apps: it drafts documents, builds spreadsheets, edits code, runs jobs on a schedule and picks up new abilities through plugins. It's in public preview, free to download, and everything below is covered step by step in our guides.

Getting set up

1
Download & install
Grab the build for your OS from the download page and open it like any desktop app.
2
Sign in or add a key
Use your DeepSeek account, or paste an API key if you prefer pay-as-you-go.
3
Open a folder
Point Harness at a project folder. It only touches the files you give it access to.
4
Start a session
Describe the task in plain language and review each change before it's applied.

Full Harness setup guide

Documents

Draft, rewrite and format reports, letters and notes; it reads your existing files for context and exports to Word or PDF.

Documents guide →

Spreadsheets

Clean messy data, write formulas, build pivots and charts, and explain what a sheet is doing in Excel or CSV files.

Spreadsheets guide →

Coding

Explain, fix, refactor or extend a codebase, run tests, and review diffs before anything is committed.

Coding guide →

Scheduled tasks

Set a job to run hourly, daily or weekly reports, inbox summaries, data refreshes and get the result without lifting a finger.

Scheduled tasks guide →

Plugins

Add connectors, skills and tools or write your own to extend what the agent can reach and do.

Plugins guide →

✦ How to use DeepSeek AI

How do I start a chat, write a prompt, upload a file, or work through a coding task?

Four things cover almost everything people do in DeepSeek. Each step below shows what the screen looks like, what to type, and the kind of answer to expect. The full walkthrough, with more examples, is on the How to Use page.

Step 1

Start a chat

Open the app or the web version and you land on an empty conversation. There's a single text box, two toggles a deeper-thinking mode for hard questions and web search for anything recent and a send button. Type, hit Enter, done. Every new topic deserves a fresh chat; it keeps answers focused.

Keep one topic per conversation.
Switch on web search for news, prices or anything after the model's training date.
Turn on the thinking mode for maths, logic and planning.

Starting your first chat →

Step 2

Write a good prompt

The quality of the answer tracks the quality of the ask. A strong prompt says what you want, who it's for, how long it should be and what tone to use. The example shows the pattern: task + audience + constraints. Paste the source material straight into the box, or attach it (step 3).

Try this Explain [topic] to a [audience] in [length]. Use [tone]. Include [one concrete example].

Prompting guide with 20 templates →

Step 3

Upload a file

Click the paperclip or drag a file onto the chat. DeepSeek reads PDFs, Word documents, spreadsheets, plain text and images, then answers questions about them. Ask for a summary, a table of the numbers, a translation, or "what's missing" and refer back to the file in later messages without re-uploading.

Example output

Key obligations: deliver v1 by 30 Nov; 30-day payment terms; 12-month support.
Flagged: clause 9.2 unlimited liability for indirect losses. Consider capping at contract value.

Working with files →

Step 4

Work through a coding task

Paste the code (or the error) and say what you expected to happen. DeepSeek explains the cause, proposes a fix as a diff, and can write the test that proves it. For larger changes across many files, move the same conversation into Harness Coding, where it edits the files and runs the tests for you.

Try this Here's my function and the failing test. Explain why it fails, then give me a minimal fix as a diff.

Coding with DeepSeek →

✦ DeepSeek API

How do I create an API key and send my first request?

The API is OpenAI-compatible, so most existing client libraries work with a change of base URL. Three steps get you from zero to a working call: create a key, set it in your environment, send a chat request. Full reference, limits and model names are on the API guide; costs per token on the pricing page.

1
Open the API platform
Sign in to the developer console (separate from the chat app) and go to API keys. Add a small credit balance if your account is new.
2
Create and copy a key
Click Create new key, give it a name you'll recognise, and copy the value immediately it's only displayed once. One key per project makes revoking easy later.
3
Store it safely
Put it in an environment variable (DEEPSEEK_API_KEY) or a secrets manager. Never commit it to a repo or paste it into client-side code.
4
Send your first request
Use the examples on the right. Point your client at the DeepSeek base URL, pick a model, and send a chat message.

# pip install openai
from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["DEEPSEEK_API_KEY"],
    base_url="https://api.deepseek.com",
)

resp = client.chat.completions.create(
    model="deepseek-chat",
    messages=[
        {"role": "system", "content": "You are a concise assistant."},
        {"role": "user",   "content": "Give me three names for a tea shop."},
    ],
)

print(resp.choices[0].message.content)

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DEEPSEEK_API_KEY" \
  -d '{
    "model": "deepseek-chat",
    "messages": [
      {"role": "system", "content": "You are a concise assistant."},
      {"role": "user",   "content": "Give me three names for a tea shop."}
    ]
  }'

// npm install openai
import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.DEEPSEEK_API_KEY,
  baseURL: "https://api.deepseek.com",
});

const resp = await client.chat.completions.create({
  model: "deepseek-chat",
  messages: [
    { role: "system", content: "You are a concise assistant." },
    { role: "user",   content: "Give me three names for a tea shop." },
  ],
});

console.log(resp.choices[0].message.content);

200 OK Sample response

{
  "id": "chatcmpl-7f3a…",
  "object": "chat.completion",
  "model": "deepseek-chat",
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": "1. Steep & Co.\n2. The Kettle Room\n3. Leaf Lane"
    },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 27,
    "completion_tokens": 19,
    "total_tokens": 46
  }
}

The usage block is what you're billed on see API pricing. For reasoning or web-grounded answers, swap the model or add tools as shown in the search & fetch guide.

✦ DeepSeek app download

Get DeepSeek on every device

Free chat on your phone, the full experience in the browser, and the Harness agent on your desktop. Pick your platform below each link goes to our download page, which lists the official store and installer links and shows you how to check you've got the real app.

iOS

iPhone & iPad · App Store

Download for iOS →

Android

Phones & tablets · Google Play

Download for Android →

Web

No install · any browser

Open in the browser →

Desktop

Harness · macOS, Windows, Linux

Download Harness →

All download links Free to use · official store links only · see download page for version notes

✦ DeepSeek vs ChatGPT

DeepSeek vs ChatGPT: which one should you use?

Both will answer questions, write, summarise and code well. The differences are in cost, openness, how the reasoning is shown, and the ecosystem around each. Here's the short version the full comparison has benchmarks, prompts we ran on both, and screenshots.

At a glance	DeepSeek	ChatGPT
Free tier	Full chat with current models, including thinking mode and web search.	Free tier with usage limits; best models and higher limits need a paid plan.
Open-weight models	✓ Most models published with weights you can self-host or fine-tune.	◐ A few open-weight releases; the flagship ChatGPT models are closed.
Visible reasoning	✓ R1 and the thinking mode show the full chain of thought.	◐ Shows a summary of its reasoning rather than the full trace.
API cost	Among the cheapest per token; OpenAI-compatible so switching is a base-URL change. See pricing.	Higher per-token prices on comparable models; broad model range.
Ecosystem	Growing: apps, API, Harness Desktop with plugins; strong community around open models.	Mature: custom GPTs, voice, image generation, deep integrations with many third-party tools.
Coding	Very strong, especially with Coder and R1 on hard problems; Harness handles whole-repo work.	Very strong; mature agentic coding tools and IDE integrations.
Data & hosting	Hosted service (check the privacy policy), or run the open weights entirely on your own hardware.	Hosted only; enterprise plans add data controls.

Choose DeepSeek if…

you want strong results at low cost, need to see the model's reasoning, or want open weights you can run yourself and build on without vendor lock-in.

Choose ChatGPT if…

you rely on its wider ecosystem voice, images, custom GPTs, plug-ins for tools you already use and prefer a single polished product over flexibility.

Read the full comparison V4 Pro vs V4.1 Flash

✦ DeepSeek login & troubleshooting

Can't sign in, or something's not working?

Most problems come down to one of a handful of causes: a sign-in method mix-up, a verification code that never arrives, heavy traffic on the servers, or a browser extension getting in the way. The checklist below fixes the common ones in a minute or two; the login guide covers every sign-in route step by step.

Ways to sign in

Email or phone + one-time code no password to remember; the code arrives by email or SMS and expires after a few minutes.
Google or Apple fastest on mobile; uses the account already on your device.
WeChat scan the QR on the web login page with the WeChat app.
API platform a separate account from the chat app; a chat login won't open the developer console.

Step-by-step login guide →

Common problems and fixes

The verification code never arrives

Wait 60 seconds and check Spam/Junk and Promotions tabs.
Make sure you typed the address or number exactly a swapped letter sends the code elsewhere.
Request a new code; the old one is void once you do.
For SMS, confirm your number has the right country code and isn't blocking short-code senders.

"Server is busy" or the reply never finishes

Demand spikes after a release. Retry in a minute, shorten the prompt, or switch off the thinking mode for simple questions. If it keeps happening, the updates page notes any known outages. The API has separate capacity, so developers are usually unaffected.

Almost always an extension or stale cookies. Try a private/incognito window first; if that works, clear site data for the DeepSeek domain and disable ad-blockers or privacy extensions for it. Check that your system clock is correct a wrong time breaks the sign-in token.

The mobile app crashes or won't update

Update to the latest version from the store, then force-quit and reopen. If it still crashes, reinstall your chats are stored on your account, not the phone. Only install from the official store links on our download page; look-alike apps are a common cause of "login failed" reports.

API returns 401 or "insufficient balance"

401: the key is wrong, revoked, or you're sending it to the chat site instead of api.deepseek.com. Check the Authorization: Bearer header and that there's no trailing space. Balance: the API is prepaid top up in the console. See the API guide and pricing.

My chat history disappeared

Check you're signed in with the same method as before email, Google and Apple create separate accounts if the addresses differ. On the web, history lives in the left sidebar; on mobile, swipe from the left edge. Deleted chats can't be recovered.

Still stuck? Contact us with a screenshot and what you tried we'll point you to the right fix.

✦ DeepSeek news & updates

What's new in DeepSeek

Release notes in plain English: what changed, who it affects, and whether you need to do anything. We update this list with every model, app and API release.

All updates →

Model releaseOctober 2026

DeepSeek V4.1 Flash updated to V4.1-Flash-1017

Significantly stronger project development, lower cost per task and faster responses. The API now supports web search and fetch, so answers can be grounded in live pages without extra plumbing. Chat users get it automatically; API users keep the same model name.

Read the release notes →

API

Web search & fetch tools arrive in the API

Two new tool types let a request search the web and pull a page's content before answering. Opt-in per request; billed as tokens.
Guide

V4 Pro vs V4.1 Flash which should you use now?

We re-ran our comparison prompts on the 1017 build. Flash closes most of the gap on everyday tasks; Pro still wins on long, nuanced work.
Harness

Harness Desktop in public preview worldwide

The open-source desktop agent is available on macOS, Windows and Linux, with plugins for documents, spreadsheets, coding and scheduled tasks.
Guide

API pricing explained, with worked examples

What a token is, how caching lowers the bill, and what a typical chatbot or summariser costs per month on each model.

Want release alerts? Bookmark the updates page or follow the feed from there.

✦ FAQ

Frequently asked questions

Quick answers to the things people ask us most. Each one links to the page with the full detail.

Is DeepSeek free to use?

Yes the chat app and web version are free, including the thinking mode and web search. The API is pay-as-you-go, billed per token; see API pricing.

Which model should I pick?

Start with V4.1 Flash for everyday tasks. Switch to R1 when you need step-by-step reasoning, Coder for pure programming, and V4 Pro when quality matters more than cost. Our model overview compares them side by side.

What's the difference between DeepSeek and ChatGPT?

Both are strong general assistants. DeepSeek is cheaper, publishes open-weight models and shows its full reasoning; ChatGPT has the larger ecosystem of integrations. Full breakdown in DeepSeek vs ChatGPT.

Can I run DeepSeek on my own computer?

Two ways. Open-weight models like R1 and Coder can be self-hosted if you have the hardware. For an agent that works on your files, install Harness Desktop, which is open source and runs on macOS, Windows and Linux.

Does DeepSeek have a mobile app?

Yes, for iOS and Android, with voice input, file uploads and web search. Only install from the official store links on our download page.

How do I get an API key?

Sign in to the API platform (a separate account from the chat app), open API keys, create one and copy it immediately. The API guide walks through your first request in Python, cURL and Node.

Is the API compatible with OpenAI's SDKs?

Yes. Point any OpenAI-compatible client at the DeepSeek base URL with your key and choose a DeepSeek model name no other code changes needed for chat completions.

What is Harness Desktop?

An open-source desktop agent that drafts documents, builds spreadsheets, edits code, runs scheduled tasks and extends itself with plugins all on your machine. See the Harness guide.

Why can't I log in?

Usually a code that hasn't arrived, a browser extension, or signing in with a different method than you registered with. Work through the login & troubleshooting checklist or the login guide.

Is this the official DeepSeek website?

No. DeepSeekR1.com is an independent guide site and isn't affiliated with DeepSeek. We link only to official downloads and never ask for your DeepSeek password. Read our About and Disclaimer pages.

Didn't find your question? Ask us and we'll add it here.

Explore DeepSeek AI.Explained. Right now.