Dataset · Qwen
Benchmark for app recreation agents
Qwen/RecreationBench
250 English test tasks to evaluate AI agents that study software and build replicas.
What's inside
The benchmark from Qwen includes 250 app recreation tasks for Ubuntu, macOS, Windows, Android, and the web. It contains English text descriptions and images taken from the held-out test split of the RecreationWorld framework. The agent receives a prompt, interacts with a running reference app, studies its behavior, and writes its own implementation.
What you can do with it
- Evaluating multimodal computer-use agents
- Testing LLM capabilities in reverse engineering and app code generation
- Tasks
- Text generation
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 46.7K
- Likes
- 19
How to load it
from datasets import load_dataset
ds = load_dataset("Qwen/RecreationBench")