Artificial intelligence continues to advance at breakneck speed, with new developments raising questions about the potential for AI to rival—and even surpass—human intelligence. The latest breakthrough comes from OpenAI’s new “o1” model, which recently underwent testing on the Norway Mensa IQ test, a popular measure of human intelligence. The results? An impressive score that may signal a major leap forward in AI reasoning capabilities.
When tested on the Norway Mensa IQ test, o1 scored 25 out of 35 questions correctly
When tested on the Norway Mensa IQ test, o1 scored 25 out of 35 questions correctly, suggesting it could have an IQ as high as 120. This result places the AI among the more intelligent humans, as an IQ of 120 is significantly above the average human IQ of 100. However, concerns were quickly raised that the AI’s performance might be inflated by training data that included IQ test questions, potentially giving it an unfair advantage.
A new test
To address these concerns, a unique experiment was designed. The researcher, Maxim Lott, worked with a Mensa member to create a new IQ test composed of entirely novel questions. These questions were never published online to ensure they weren’t part of the AI’s training data. The new test was carefully aligned with the Mensa Norway test to ensure that both measured intelligence in a similar way.
When given this newly designed test, the o1 model’s score was closer to that of the average human, landing around an IQ of 100. Although this was lower than its earlier score, it still demonstrated a level of intelligence equivalent to that of an average person.
In a detailed blog post, Lott shared his thoughts on the implications of these results. While he initially believed AI progress had plateaued after seeing no improvement since February 2024, the recent results showed a sudden and significant leap. He explained that o1’s high performance on the Norway Mensa test, followed by its respectable performance on the new test, indicates that the AI’s reasoning ability has genuinely improved—not just its ability to recall answers from training data.
“The magnitude of o1’s lead is still huge,” Lott noted, emphasizing that these results are evidence of real progress in AI’s ability to reason and solve problems. He also highlighted that further research could refine the precision of these scores, particularly by offering multiple wordings of test questions to see how AIs respond to different descriptions.
By 2026, AI models could potentially achieve IQ levels as high as 140
The broader implications of these developments are striking. If AI models continue to improve at this rate, they may soon surpass human intelligence. Lott outlined a timeline for future AI progress, predicting that by 2026, we could see AI models reaching IQ levels of 140 or higher, a milestone that would put them among the smartest humans on Earth. He also mentioned the possibility of AI eventually surpassing all human intelligence within 4 to 10 years.
Lott’s analysis aligns with recent trends in AI development. He pointed to the progression of other models, such as Claude, which showed consistent IQ improvement in recent years. OpenAI’s o1 model now seems to be leading the pack, positioning the company ahead of competitors like Anthropic in the race to develop the world’s most intelligent AI.
What’s next for AI intelligence?
As we look toward the future, the rapid advancements in AI are both exciting and concerning. The prospect of AI surpassing human intelligence could revolutionize industries and reshape society. Yet, it also raises questions about the long-term implications for humanity. Will AI become a tool that enhances human capabilities, or will it eventually become something more, outpacing us in ways we have yet to fully understand?
For now, the o1 model’s performance offers a glimpse into the potential of AI’s cognitive abilities. But as Lott warns, this is just the beginning. “Hold onto your hats,” he advises because the future of AI is arriving faster than we might have expected.