Claude Fable 5.1 and GPT-6 Astra have achieved a flawless score of 151 points—the highest possible score—seven consecutive times on the Mensa Norway IQ test. This test comprises 35 visual reasoning questions to be answered within a 25-minute timeframe. Such a score significantly surpasses the 130-point threshold required for Mensa membership, positioning these AI systems within the upper echelons of human intellectual capability. The TrackingAI leaderboard reveals that several leading AI models have exceeded 140 points on this test, with domestic models also ranking among the top performers. Visual reasoning questions are designed to assess human fluid intelligence, an area that was traditionally considered a challenge for AI systems. In early 2024, most AI models scored around 64 points on these tests. However, the scores of leading AI models have rapidly improved, with an average monthly increase of 2.5 points from May 2024 to October 2025, a rate of progress far exceeding the Flynn effect observed in human IQ growth over time.
To alleviate concerns that AI systems might simply be memorizing answers, test administrators employed new, confidential versions of the test. Despite these measures, leading AI models still scored close to perfect marks, demonstrating a significant enhancement in their reasoning abilities. Currently, the pool of challenging problems designed by humans to differentiate AI capabilities is being rapidly depleted. This situation has prompted researchers in related fields to revise their estimates for the arrival of Artificial General Intelligence (AGI) forward. The longstanding premise of IQ tests, which humans have historically used to assert their intellectual superiority, is now facing scrutiny and reevaluation.