// AI detectors
Tencent Zhuque AI Detector
Free AI-generated text and image/video detection from Tencent, with suspicious paragraphs highlighted.
official sources onlychecked 2026-10-26
01At a glance
| Official link | https://matrix.tencent.com/ai-detect/ |
|---|---|
| Made by | Tencent（腾讯朱雀实验室） |
| Free? | Free. Guests: 3 text + 3 image checks per day. Logged in: 20 text + 30 image/video checks per day; feedback (like/dislike) adds 5 more, up to 30 bonus. |
| Paid | API for businesses via Tencent Cloud EdgeOne “Zhuque AIGC Detection Model” (pricing on Tencent Cloud). |
| Languages | Chinese and English (benchmarks published for both). |
| Files | Text: txt, docx; images: png, jpg, webp; video: mp4, mov, avi. |
| Limits | Per-check character limit is not stated on the official page (check the official page). |
| Best for | Students, teachers and editors checking Chinese text; media checking AI images. |
| API | Yes – Tencent Cloud EdgeOne, models zhuque-text and zhuque-image, sync and async calls. |
02How to use it
- Open matrix.tencent.com/ai-detect and choose Text or Image/Video.
- Paste text or upload a file (txt/docx, or png/jpg/webp/mp4/mov/avi).
- Log in for more daily checks; read the highlighted paragraphs as hints, not proof.
03Accuracy
Official: overall text accuracy above 92%; image accuracy 98%+. Official benchmark table (data updated 2026-06-30): HC3 Chinese 99.00%, Detect-RL Chinese 98%, AIGTBENCH 96.9%, MAGE 97.72%, RAID TPR@FPR=5% 99.95%, HC3-en 100%, M4GT 100%. Model updated 2026-07-21.
Independent evaluations:
- Weber-Wulff et al. (2023), “Testing of detection tools for AI-generated text”, International Journal for Educational Integrity — 14 tools tested (12 public + Turnitin and PlagiarismCheck): all scored below 80% accuracy and only 5 above 70%; the authors conclude the tools are “neither accurate nor reliable”. False positives ranged from 0% (Turnitin) to 50% (GPTZero) on their test set; machine paraphrasing cut accuracy to 26%. English texts, tested in 2023 – tools have changed since.
- Liang et al. (2023), “GPT detectors are biased against non-native English writers”, Patterns — Several widely used GPT detectors “consistently misclassify non-native English writing samples as AI-generated, whereas native writing samples are accurately identified”; simple prompting can also bypass them. The authors caution against using detectors in evaluative or educational settings.
04FAQ
FAQ
Is Zhuque AI detection free?
Yes. Guests get 3 text and 3 image checks a day; logged-in users 20 text and 30 image/video checks a day, plus up to 30 bonus checks for feedback.
How accurate is Zhuque?
Tencent reports overall text accuracy above 92% and 98%+ for images, with benchmark results on its page – but it also states the tool can produce false positives. AI detection may misjudge; do not treat a score as proof.
What files does Zhuque support?
txt and docx for text; png, jpg and webp for images; mp4, mov and avi for video.
Can Tencent Zhuque AI Detector be wrong?
Yes – AI detection can misjudge. Tencent itself says Zhuque “is not 100% accurate and can have false positives” – e.g. web novels and elementary-school essays written in simple, generic language – and results should be treated as reference only.
05Official sources
Checked against these pages on 2026-10-26; the official page wins if anything differs.