Medli1108/Open_CaptchaWorld

---
language:
- en
license: apache-2.0
size_categories:
- n<1K
task_categories:
- visual-document-retrieval
tags:
- Multimodal_Agents
- Open_Source_CAPTCHAs
---

# Open CaptchaWorld Dataset

This dataset accompanies the paper [Open CaptchaWorld: A Comprehensive Web-based Platform for Testing and Benchmarking Multimodal LLM Agents](https://huggingface.co/papers/2505.24878). It contains 20 distinct CAPTCHA types, each testing different visual reasoning capabilities. The dataset is designed for evaluating the visual reasoning and interaction capabilities of Multimodal Large Language Model (MLLM)-powered agents.

[Project Page](https://huggingface.co/spaces/OpenCaptchaWorld/platform) | [Github](https://github.com/Yaxin9Luo/Open_CaptchaWorld)

The dataset includes:

* **20 CAPTCHA Types**: A diverse set of visual puzzles testing various capabilities. See the Github repository for a full list.
* **Web Interface**: A clean, intuitive interface for human or AI interaction.
* **API Endpoints**: Programmatic access to puzzles and verification.

This dataset is useful for benchmarking and improving multimodal AI agents' performance on CAPTCHA-like challenges, a crucial step in deploying web agents for real-world tasks. The data is structured for easy integration into research and development pipelines.