QwenCloud's official vision API documentation demonstrates image-question-answering requests using the official model ID qwen4-max. Images are processed at high detail by default; you can set vl_high_resolution_images or max_pixels to control visual resolution and token usage.
Suitable tasks: Image question answering, reading small text, identifying local details, and extracting text from images
Unsuitable tasks: Tasks requiring guaranteed pixel-level recognition accuracy, or treating the example as a complete list of image format and size limits
Applicable model version: qwen4-max in the official example
Applicable client, agent, or API: QwenCloud OpenAI-compatible Chat Completions; the example uses the OpenAI Python SDK
Recommended reasoning tier and parameters: detail=high by default; enable high-resolution mode for reading details, or set max_pixels for the task
The following example uses the official Singapore-region endpoint and request structure. First, set the API key for the matching region:
export DASHSCOPE_API_KEY="YOUR_QWENCLOUD_API_KEY"import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["DASHSCOPE_API_KEY"],
base_url="https://ap-southeast-1.maas.qwencloudapi.com/compatible-mode/v1",
)
response = client.chat.completions.create(
model="qwen4-max",
messages=[{
"role": "user",
"content": [
{
"type": "image_url",
"image_url": {
"url": "https://example.com/image.jpg",
# 可选:设置单图像素上限;Qwen4-Max 最大为 16,777,216。
# "max_pixels": 8_000_000,
},
},
{"type": "text", "text": "请读出图片右下角的编号,并说明不确定之处。"},
],
}],
# 可选:保留更高分辨率的图像细节,视觉 token 消耗也会增加。
# extra_body={"vl_high_resolution_images": True},
)
print(response.choices[0].message.content)Choose one configuration based on the level of detail required: the default configuration costs less; enable vl_high_resolution_images=True when reading small text or local details. To set a pixel limit explicitly, use max_pixels; the limit for Qwen4-Max is 16,777,216.
Use the endpoint for the same region as the API key, and set model to qwen4-max.
Pass a publicly accessible image URL through image_url, and include a text instruction in the same user message.
To read small text or details, enable vl_high_resolution_images or increase max_pixels; observe changes in input tokens and results.
If the request fails, check image URL accessibility, request size, format, and resolution limits.
The documentation's vision API example uses model="qwen4-max" in OpenAI-compatible Chat Completions and includes both single-image and multi-image request examples, confirming that this general-availability model version supports visual input.
The documentation states that Qwen4-Max defaults to detail=high. Without max_pixels, high processes up to 16,777,216 pixels, while low and auto are capped at 2,621,440 pixels.
Qwen4-Max uses a configuration of 32 × 32 pixels per visual token, with a visual-token limit of 16,384. The configurable max_pixels defaults to 2,621,440 and has a maximum of 16,777,216.
The shortest image dimension must be greater than 10 pixels, and the aspect ratio must not exceed 200:1. The recommended maximum is 8K (7680 × 4320); higher resolutions may increase transfer time and cause timeouts.
Below 8K, BMP, JPEG, PNG, TIFF, WEBP, and HEIC are supported; from 4K to 8K, only JPEG, JPG, and PNG are supported. The total size limit per request is 64 MB; for Qwen4-Max, the limit for a single image from a public URL and for the original image before Base64 encoding is 20 MB.
vl_high_resolution_images increases visual-token usage. The official documentation says Qwen4-Max can still use an explicit max_pixels in this mode; if none is set, images are processed at up to 16,777,216 pixels.
Increasing max_pixels does not guarantee accurate results. The official recommendation is to resize ultra-high-resolution images to a reasonable size first; outputs for detail recognition should still be checked.
The format, pixel, and size limits above come from the current QwenCloud vision documentation; other models or regions may have different limits. The endpoint example is for the Singapore region. When changing regions, use both that region's endpoint and API key.
This article focuses on image input and resolution. For API keys, SDK installation, and general integration steps, see the QwenCloud Codex integration configuration in the same directory.
Opened and checked QwenCloud's official vision documentation through Tabbit on 2026-11-17, reviewing the qwen4-max image request example, resolution parameters, and image limits. The example was adapted from the official OpenAI-compatible interface; no request was made with a real API key.
Qwen4-Max