API for Qwen: connecting Qwen 3.8 Max, pricing and Python examples

If you need an API for Qwen, the shortest path looks like this: a key from the dashboard, the OpenAI SDK and the qwen3.8-max model. A separate account with the maker's cloud and a special SDK are not required: the request format is the same as chat.completions. Below are the connection, examples with images and a long context, and a price calculation.

What qwen3.8-max is in our table

Figures as of 18 October 2026 (prices in dollars per 1M tokens):

The strength index is the Intelligence Index from Artificial Analysis; we give no other quality figures. Test whether a model fits your task on your own data. For comparison: the official prices for qwen3.8-max in our table are 2 / 6. The current list of models and prices: GET https://api.zukko.pro/v1/models or the "Models" page in the dashboard.

How to get a Qwen API key

People often ask about a "qwen ai api key". Here it is a single gateway key of the form fetch_... that works for all models, not only Qwen.

1. Open the dashboard↗ and top up the balance (from $1: SBP, cryptocurrency or MasterCard cards).
2. In the API keys↗ section, create a key. The full key is shown once.
3. If you like, restrict it: a daily and monthly dollar limit and a list of allowed models (allowed_models), for example only qwen3.8-max.

It is better to keep the key in an environment variable rather than in code.

The first request in Python

pip install openai
export ZUKKO_API_KEY=fetch_your_key

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.zukko.pro/v1",
    api_key=os.environ["ZUKKO_API_KEY"],
)

r = client.chat.completions.create(
    model="qwen3.8-max",
    messages=[
        {"role": "system", "content": "You are an assistant. Answer briefly."},
        {"role": "user", "content": "Explain request idempotency in three sentences."},
    ],
    max_tokens=400,
)
print(r.choices[0].message.content)
print(r.usage)

The trailing /v1 in the address is mandatory. More Python examples and error handling are in the Python reference↗.

The same request via curl

curl https://api.zukko.pro/v1/chat/completions \
  -H "Authorization: Bearer $ZUKKO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-max",
    "messages": [{"role": "user", "content": "Hello!"}],
    "max_tokens": 100
  }'

Streaming

stream = client.chat.completions.create(
    model="qwen3.8-max",
    messages=[{"role": "user", "content": "Write a short description of a delivery service"}],
    stream=True,
)
for chunk in stream:
    if chunk.choices:
        delta = chunk.choices[0].delta.content
        if delta:
            print(delta, end="", flush=True)

The if chunk.choices: check is needed because the last chunk, which carries usage data, arrives with an empty choices list.

Working with images

qwen3.8-max accepts images as input (qwen3.8-flash does not). An image can be passed as a data URL:

import base64

with open("receipt.jpg", "rb") as f:
    b64 = base64.b64encode(f.read()).decode("ascii")

r = client.chat.completions.create(
    model="qwen3.8-max",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "List the receipt items and the total as JSON."},
            {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{b64}"}},
        ],
    }],
    max_tokens=600,
)
print(r.choices[0].message.content)

How many tokens an image takes depends on its size, so look at the usage field and measure on several files. Oversized images may return 413: shrink them before sending. More on recognition in the post on receipts and screenshots↗.

Long context

The stated context of the model in the table is 1M tokens. That does not mean every request should be huge: input is paid each time. If you put 500,000 tokens into a request (a large contract or a chunk of a codebase) and get a 2000-token answer:

500,000 × 0.088 / 1,000,000 = $0.044 2000 × 0.238 / 1,000,000 = $0.000476 Total about $0.0445.

At the official prices from our table (2 / 6) the same request looks like this: 500,000 × 2 / 1,000,000 = $1.00 and 2000 × 6 / 1,000,000 = $0.012, total $1.012. That is enough for comparing orders of magnitude. But always remember that with repeated questions on one document the input is paid again, and for qwen3.8-max the cache read price (0.088) equals the input price, so the cache gives no discount. Long contexts in general are covered in the post on the million-token context window↗.

The price of a typical request

Suppose 3000 input tokens (a question and a text fragment) and 600 output tokens:

3000 × 0.088 / 1,000,000 = $0.000264 600 × 0.238 / 1,000,000 = $0.0001428 Total $0.0004068, about $0.0004.

A thousand such requests cost about $0.41 (by the formula: 1000 × 0.0004068).

Which Qwen model to choose

• qwen3.8-max: you need image input or a higher strength index (45 versus 40).
• qwen3.8-flash: mass text tasks where price matters (0.038 / 0.075). More in the post on Qwen 3.8 Flash↗.

Which is better for your task is shown by running 20 real examples on two models. We make no claims about quality.

Typical errors

• 401: wrong key.
• 402: not enough balance.
• 403, model_not_allowed_for_key: this model is not allowed in the key.
• 404, model_not_found: a typo in the model name (the correct one is qwen3.8-max).
• 413: the request is too large, for example an uncompressed image.
• 429: a key limit or a limit at the model provider.
• 503: the provider is unavailable or the model is disabled (model_disabled): choose another model from GET /v1/models.

Always check exact model names and prices against GET /v1/models.

FAQ

Do I need a separate key from Alibaba or another cloud?

No. To call qwen3.8-max through the gateway you use a single fetch_... key from the dashboard.

Does qwen3.8-max support images?

Yes, qwen3.8-max accepts image input. qwen3.8-flash has no image input.

How much does a thousand requests cost?

For a request of 3000 input and 600 output tokens, about $0.41 per thousand. The real figure is shown by the usage field and the "Usage" page in the dashboard.

Can I call Qwen in the Anthropic format?

Any model can be called in both the OpenAI format and the Anthropic format. For Qwen the OpenAI format with base_url ending in /v1 is usually more convenient.