SOTA AI Text Detection benchmarks and papers with code

Benchmarks

Dataset Name	SOTA Method	Metric
HC3 AI-Prob		Finance Accuracy84.1	24	2mo ago
Unified RL corpus (test)	TELL	AUROC92.7	16	6d ago
Modern RAID Claude generated 200 tokens	DMAP	TPR@0.1%37.34	14	26d ago
Modern RAID Gemini generated 200 tokens	DALD	TPR@0.1% FPR34.74	14	26d ago
Modern RAID GPT-5 generated 200 tokens	DALD	TPR@0.1%30.22	14	26d ago
AI-Generated Text (Held-out Set)	PASTA	AI-detection Rate (Aligned, Pangram)100	11	2d ago
MELD (eval)	MELD	AUROC99.99	10	23d ago
Ghostbuster	MELD	AUROC100	10	23d ago
M4GT	MELD	AUROC78	10	23d ago
HC3	RoBERTa-ChatGPT	AUROC100	10	23d ago
RAID No attack clean (test)		AUROC99.97	9	23d ago
RAID All settings with attack (test)		AUROC99.87	8	23d ago
RAID Mistral 1.0 (test)	AEyeDE	Accuracy (Acc)95.09	7	1d ago
RAID Llama 1.0 (test)	AEyeDE	Accuracy98.99	7	1d ago
RAID GPT-neo 1.0 (test)	AEyeDE	Accuracy97.22	7	1d ago
RAID Cohere 1.0 (test)	AEyeDE	Accuracy97.23	7	1d ago
Product Reviews	GPTZero	AUC99.9	7	3mo ago
Paper Reviews	GPTZero	AUC0.999	7	3mo ago
Creative Writing	GPTZero	AUC99.9	7	3mo ago
Abstracts	GPTZero	AUC99.9	7	3mo ago
Modern RAID	DALD	TPR @ FPR=0.1%28.44	6	26d ago
Peer-Review	Calibrated DMAP	TPR@0.1%67.36	6	26d ago
M4 (macro-average)	DeBERTa-v3+FeatAttn	Balanced Accuracy (BA)84.5	6	28d ago
electra v1 (representative cell)	Signed-ε rank-1 intervention	Cp TPR100	5	12d ago
Frankentext Generations 1.0 (test)	Frankentext (Qwen-3-32B thinking)	Pangram AI %85	5	1mo ago

Showing 25 of 43 rows