AI Hallucination Rankings Unveil Surprising Model Performances

The AI Hallucination Leaderboard from Vectara shows that Google’s model performs exceptionally well, and OpenAI’s o3-mini-high also achieved solid results. Notably, China’s open-source model GLM from Zhipu AI ranked impressively as well. The leaderboard evaluates hallucination by asking large language models questions related to specific articles, then assessing their responses using Vectara’s own evaluation model. The results have sparked widespread discussion, with users analyzing and providing feedback on the performance of various models—particularly focusing on those that were not included in the evaluation—and expressing hopes for improvements in future iterations.