Open Detector
BERT classifier that scores whether academic text reads as human or AI-written, run locally with Python.
A BERT model on Hugging Face that labels academic paragraphs as human or AI style.
Open Detector is a BERT-base binary classifier for telling human academic writing from AI-written style. It suits students, researchers and technical users who want a local, inspectable alternative to commercial paper checkers.
What it is
The repo provides a model hosted on Hugging Face (followsci/bert-ai-text-detector) and a short Python snippet that returns AI and human percentages for a paragraph. The README says it is stylometric detection only, not a check of content truth.
Who it's for
It suits people checking academic prose who can run Python. It does not offer a web interface or sentence highlighting in the repository, and the README points to a separate hosted platform for those.
Key features
BERT-base classifier
Binary human-versus-AI model fine-tuned on about 1.4 million samples, per the README.
Academic text focus
Trained and described for academic writing patterns, not general web copy.
Hugging Face model
Weights are downloaded from followsci/bert-ai-text-detector; the model card states MIT.
Short Python quick start
pip install transformers torch, then score a paragraph in a few lines.
Reported false-positive rate
README claims a 0.82% false-positive rate on its own dataset.
English and Chinese docs
README is provided in English and Chinese.
Up and running in 3 steps
Our honest verdict
What's great
- Tiny setup: two pip packages and a short script.
- MIT licence on the code and the model card.
- Runs locally, so paragraphs are not sent to a vendor.
- The README states its limits, including that it only measures writing style.
Where it falls short
- Input is truncated at 512 tokens, so long papers need splitting.
- Accuracy figures such as 99.57% come from the authors' own dataset, not independent data.
- No web interface, document upload or highlighting in the repo itself.
- Tuned for academic text; results on emails or marketing copy are not documented.
Pricing & license
Open Detector is free under the MIT licence. The README also links a separate hosted platform, XYZ SCIENCE, with free detection; otherwise the only cost is running the model yourself.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Is Open Detector a Turnitin replacement?
The README positions it as an open alternative to commercial detectors such as Turnitin for AI-style detection, but it does no plagiarism matching.
Can it run offline?
Yes after the model is downloaded; the quick start uses only transformers and torch on your own machine.
Does it prove a paper was written by AI?
No. The README calls it stylometric detection and argues that writing like AI is not misconduct.
