// the find
XiaomingX/openai-captcha-detection
openai-captcha-detection 是一个使用 OpenAI 进行验证码识别的工具。目前验证码识别准确率100%，通过调用 OpenAI 的 API，这个项目可以实现对复杂验证码图片的文本识别，帮助开发者在验证码处理场景中进行自动化操作。openai-captcha-detection is a tool that uses OpenAI for captcha recognition. Currently, it achieves a 100% accuracy rate in captcha recognition.
A thin wrapper around GPT-4 Vision that sends captcha images to OpenAI and returns the text. It's essentially a ~50-line script dressed up as a library. The claimed 100% accuracy rate is based on exactly one test image in the repo.
- The core idea is sound — GPT-4V handles distorted text better than traditional OCR models for many captcha styles
- Includes a local OCR fallback demo (local_ocr_demo.py) alongside the API version, which is a useful comparison point
- MIT licensed, no strings attached
- The '100% accuracy' claim is marketing noise — there is one sample image in the repo and no benchmark dataset, no test suite, nothing
- No batching, no rate limiting, no retry logic — calling this in a loop will hit OpenAI rate limits immediately with no graceful handling
- This is a demo script, not a library: no packaging, no installable module, imports assume you're running from inside the src/ directory
- The author's own README links this project as a security/hacking tool alongside a CVE-PoC collector and .git leak scanner — the intended use case is bypassing bot protection, not building accessibility tools