SlopTotal
sloptotal.com
- Free plan
- Open source
- Windows
- macOS
- Linux
- Web
- Browser extension
- API
- Self-hosted
8.5 Club score Visit sloptotal.com
Plans
Every plan and priceFree
Free
Free plan
What the maker publishes
- API access
- The self-hosted API requires neither an account nor an API key. source
- API endpoints
- Documented endpoints include full analysis, quick scoring, paragraph scoring, site scanning, file extraction and report retrieval. source
- Benchmark AUC
- The published held-out benchmark reports an AUC of 0.942. source
- Benchmark size
- SlopBench contains 3,252 texts split evenly between pre-ChatGPT human writing and outputs from 14 AI models. source
- Browser extension
- The SlopTotal Chrome extension can mark AI-looking results in Google Search and LinkedIn. source
- Chrome extension
- The extension can mark AI-looking content in Google Search and LinkedIn and can point to a selected SlopTotal backend. source
- Code limitation
- SlopTotal states that its natural-language engines are not intended to detect AI-written source code. source
- Detection methods
- Its 23 engines include neural classifiers, statistical tests and linguistic heuristics that run in parallel. source
- Detection methods
- The engine set combines neural classifiers, GPT-2-based statistical tests and linguistic heuristics. source
- Edited-text limitation
- Humanized AI text was detected at a 55.4% rate in the benchmark's humanizing-prompt slice. source
- False positives
- At the Likely AI threshold, 1.7% of benchmark human texts were classified as AI. source
- GPU support
- CUDA is optional for faster inference; a GPU is not required. source
- Hosted API
- The hosted service is available at api.sloptotal.com and uses a shared queue with 30-day report retention. source
- Human review
- The maker says a detector score is evidence rather than a verdict and should not by itself decide a grade, hire or publication. source
- Input formats
- The repository documents text, URLs and extraction from PDF, DOCX, TXT and Markdown files. source
- Licence
- The application code is released under the MIT licence, while model weights retain their own licences. source
- Limit: human misclassification example
- Korben reports a 2020 human blog post scoring 56.9 and receiving a Likely AI-generated verdict. source
- Limit: independent test false positives
- Korben's 36-text test classified 4 of 12 human samples as AI, all in English. source
- Limit: independent test misses
- In the same test, 9 of 24 AI-written samples were judged clean, all in French. source
- Limit: sample result
- Korben reports a 455-word Claude article scoring 23.6 and being labelled clean. source
- Maker profile
- The linked GitHub profile identifies the maker as Pablo Carrasco, a software engineer and computer-science graduate located in Valencia, Spain. source
- Non-native English limitation
- 15.4% of TOEFL essays by non-native English writers crossed the Likely AI threshold in the published evaluation. source
- Open source and self-hosting
- The detection backend is open source and can be self-hosted on the user’s own hardware. source
- Per-engine transparency
- Reports display each engine's score alongside the combined result. source
Checked against the maker's pages on 15 Oct 2026.