OpenAI launched its o3 and o4 mini reasoning models, claiming they approach AGI. However, a report reveals these models exhibit increased hallucinations, with o3 hallucinating 33% of the time, compared to lower rates in previous models. More research is needed to determine the underlying causes.