Hallucination Detection on PubmedQA benchmark leaderboard · SOTA2 Research

Hallucination Detection on PubmedQA

88F1 ScoreLatentAudit

Evaluation Results

Method	Links
LatentAudit2026.04		88	—	94.8
GPT-4o Judge2026.04		87	—	94.8
GPT-4o Judge2026.04		87	—	94.8
GPT-4o Judge2026.04		87	—	94.8
GPT-4o Judge2026.04		87	—	94.8
GPT-4o Judge2026.04		87	—	94.8
LatentAudit2026.04		87	—	94.2
LatentAudit2026.04		86	—	93.1
LatentAudit2026.04		86	—	93.8
LatentAudit2026.04		85	—	92.5
INSIDE2026.04		84	—	90.8
INSIDE2026.04		83	—	90.3
INSIDE2026.04		83	—	90.5
INSIDE2026.04		82	—	89.9
INSIDE2026.04		82	—	89.5
SAPLMA2026.04		82	—	88.2
HalluClean2025.11		81.7	81	—
SelfCheckGPT2026.04		81	—	87.1
SAPLMA2026.04		81	—	87.8
SAPLMA2026.04		81	—	88
SelfCheckGPT2026.04		80	—	86.2
SelfCheckGPT2026.04		80	—	86.9
SAPLMA2026.04		80	—	87.2
SAPLMA2026.04		80	—	87
SelfCheckGPT2026.04		79	—	85.5
SelfCheckGPT2026.04		79	—	85.8
DeepSeek-R12025.11		76.4	79	—
DeepSeek-V32025.11		71	77.5	—
Min-Perplexity2026.04		70	—	72.2
Min-Perplexity2026.04		69	—	71.8
Min-Perplexity2026.04		69	—	72
GPT-4o-mini2025.11		68.7	74.5	—
Min-Perplexity2026.04		68	—	71.5
Min-Perplexity2026.04		68	—	71.2
Llama-3-70B2025.11		16.5	54.5	—
GPT-3.5-turbo2025.11		7.7	52	—