How to access Qwen - chat, apps & API

Three ways to access Qwen

Which door you walk through depends on one question: are you here to chat, or here to build?

Qwen is Alibaba Cloud's family of large language models, and there is no single "Qwen app" you have to install before you can use it. Instead there are three distinct front doors, each suited to a different kind of person. Most people only ever need one of them, so the first job is figuring out which one is yours.

Access Qwen Chat in your browser

The fastest way to use Qwen needs no download, no install, and no payment - you can be chatting in under thirty seconds.

Qwen Chat is the official conversational interface for the Qwen models, built and maintained by Alibaba's Tongyi Lab. It is a free, web-based workspace where you can hold long multi-turn conversations, upload documents and images, generate and run code, search the live web with citations, and even produce interactive artifacts - all in the same window, with no setup required. For the overwhelming majority of people, this is the only access method they will ever need.

Desktop & mobile apps

The same account, chats, and prompts sync automatically across every platform - so you can start on your laptop and finish on your phone.

If you use Qwen often, a native app is more convenient than keeping a browser tab open. The desktop apps give you a dedicated window, keyboard shortcuts, and quicker file drops; the mobile apps add voice input and camera-based vision Q&A, so you can photograph a whiteboard or a document and ask about it on the spot. Whichever platform you choose, signing in with the same account keeps your full conversation history in sync.

Installing the desktop app is the usual three-step affair: download the installer for your operating system from the official Qwen site, run it (on macOS, drag the app into your Applications folder; on Linux, mark the AppImage executable and launch it), then sign in with your Qwen account. On mobile, search for the official Qwen app on the App Store or Google Play, or sideload the APK on Android if you prefer. Always download from the official source rather than a third-party mirror, since unofficial builds can be tampered with.

Set up the Qwen API, step by step

From zero to a working API call in about five minutes - get a key, install the SDK, point it at the right endpoint, and send your first request.

The Qwen API is delivered through Alibaba Cloud Model Studio, the unified developer platform for everything Qwen. Under the hood the API surface is called DashScope - you will see both names in the documentation, and they refer to the same service. The key thing for developers is that the API is OpenAI-compatible: it works with the standard openai Python package, the OpenAI JS SDK, LangChain, LiteLLM, and anything else that speaks the OpenAI protocol. Migrating existing code usually means changing just three things - the base URL, the API key, and the model name.

Part 1 · Get your API key

Set your key as an environment variable so you never have to hardcode it:

# macOS / Linux - current session only
export DASHSCOPE_API_KEY="sk-your-key-here"

# Make it permanent - add to ~/.bashrc or ~/.zshrc
echo 'export DASHSCOPE_API_KEY="sk-your-key-here"' >> ~/.bashrc

# Windows PowerShell
$env:DASHSCOPE_API_KEY = "sk-your-key-here"

Part 2 · Pick your endpoint

Model Studio is deployed in four regions, each with its own endpoint and its own keys. Choose the one closest to your users - or whichever your compliance team approves - and use that base URL everywhere.

Part 3 · Install the SDK

For almost everyone, the OpenAI-compatible interface is the right choice - install the official openai package and you are done. Use a virtual environment to keep things tidy:

# (optional) create and activate a virtual environment
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate

# install the OpenAI-compatible SDK
pip install openai

# OR - only if you need DashScope-specific features
# (batch invocation, advanced multimodal, real-time speech)
pip install dashscope

Your first Qwen API calls

With a key, an endpoint, and the SDK installed, every standard feature works exactly as it does with OpenAI.

Your first request

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.getenv("DASHSCOPE_API_KEY"),
    base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
)

completion = client.chat.completions.create(
    model="qwen-plus",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Who are you?"}
    ]
)

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

Streaming responses

For chatbot-style apps, stream tokens as they are generated:

stream = client.chat.completions.create(
    model="qwen-plus",
    messages=[{"role": "user", "content": "Explain async I/O simply."}],
    stream=True,
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

Vision input (multimodal)

For image tasks, switch to a vision-capable model and add an image block to the message:

response = client.chat.completions.create(
    model="qwen-vl-plus",
    messages=[{
        "role": "user",
        "content": [
            {"type": "image_url",
             "image_url": {"url": "https://example.com/chart.png"}},
            {"type": "text", "text": "Extract this chart's data as JSON."}
        ]
    }]
)
print(response.choices[0].message.content)

Choosing a model & what it costs

One of Qwen's strongest selling points is price - often an order of magnitude cheaper than GPT-4-class endpoints, with a free tier large enough to actually prototype on.

You access the whole Qwen catalogue through the same key and endpoint - you simply change the model string. A good default is a balanced general model; reach for cheaper high-volume models when latency and cost matter more than peak quality, a reasoning flagship for the hardest problems, and the specialists for code, vision, or math. Here is a snapshot of popular options on the International endpoint:

Two billing nuances are worth knowing. First, several models are priced in tiers based on the input size of each request - a short request and a near-maxed-out one fall into different rate brackets, so keeping requests lean saves money. Second, batch invocation gets a 50% discount on both input and output tokens, ideal for non-real-time workloads like overnight document processing or dataset labelling, at the cost of asynchronous (non-instant) results.

Troubleshooting common errors

Most access problems trace back to one of a small handful of causes - here's how to fix them fast.

Frequently asked questions