File size: 2,218 Bytes
ecb89cb
 
e8c6acd
 
 
ecb89cb
a711b4d
ecb89cb
 
 
 
 
 
e8c6acd
 
d8a5bbf
e8c6acd
3ab2121
0ebd454
caeec11
 
3e3cacd
e8c6acd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8085243
e8c6acd
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
---
title: Free Qwen 3.8 Max
emoji: 👀
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 6.19.0
python_version: '3.13'
app_file: app.py
pinned: false
short_description: A Space to try the new Qwen 3.8 Max for free
---

# Free Qwen 3.8 Max

Try Qwen 3.8 Max with text and image inputs. Thinking mode is enabled by default; its reasoning summary appears in a collapsible panel while the final answer streams token by token.

The Space also includes a box-prompting playground inspired by [Roboflow's Qwen 3.8 Max evaluation](https://blog.roboflow.com/qwen3-8-max/): click and drag positive and optional negative bounding boxes directly over an image, then ask Qwen to find similar objects. Three ready-made presets demonstrate shape, color, and size matching; each loads an editable instruction and example boxes. Reasoning and JSON tokens stream separately, then detection results are parsed from normalized XYXY coordinates and rendered back onto the original image.

A separate zero-shot object-detection tab accepts a plain-language detection prompt—or a **Detect everything** option—and renders class-labeled boxes without requiring examples. It includes ready-made shape, vehicle, and product-inventory prompts and uses the same streamed reasoning and JSON presentation.

This is an unofficial community demo, generously backed by a limited pool of Qwen API tokens. Availability is best-effort and the demo may be paused when the pool is exhausted.

## Configuration

The Space expects one secret:

- `DASHSCOPE_API_KEY`: a DashScope International API key with access to `qwen3.8-max`

## Local development

```bash
uv sync
DASHSCOPE_API_KEY=... uv run python app.py
```

The app limits request concurrency, image size, queue depth, and per-session request frequency to keep the public token pool useful for more people.

## Credits

- [Qwen 3.8 announcement](https://qwen.ai/blog?id=qwen3.8)
- [OpenAI-compatible vision input format](https://developers.openai.com/api/docs/guides/images-vision#analyze-images)
- [Qwen logo asset](https://logos-download.com/brands/qwen)

Qwen and DashScope are trademarks or services of their respective owners. This Space is not an official Qwen or Alibaba product.