Friday, 20 March, 2026
OpenAI's New Reasoning Models Show Increased Hallucination Rates
By Isha
OpenAI's latest reasoning AI models, o3 and o4-mini, exhibit higher hallucination rates compared to their predecessors. Internal tests revealed that o3 hallucinated in 33% of responses on the PersonQA benchmark, while o4-mini's rate was 48%, surpassing earlier models like o1 and o3-mini. The company is investigating the causes, noting that the models' tendency to generate more claims may lead to both accurate and inaccurate outputs.
Read full story at TechCrunch