For the first time since 1946, the "Mensa Club," the world's highest IQ society, is welcoming a non-human member. According to the latest LisanBench scores, GPT-5.5 Pro has achieved a text IQ of 130 (meeting the Mensa Threshold) and a visual IQ of 145, entering the "genius" tier. A year ago, the consensus in the tech industry was that "LLMs could not surpass 130." Today, that wall has been completely shattered.
This is not an isolated incident. Earlier this month, GPT-5.4 Pro scored an IQ of 150 on the TrackingAI public leaderboard. Over the past few months, OpenAI has maintained an unprecedented pace of innovation, including visual base models, reasoning upgrades, Agent product lines, and a series of API ACTions. The AI arena has officially re-entered the "openai Era."
From Weakness to Strength: The Visual Revolution
The disparity between text and visual capabilities is striking. While GPT-5.5 Pro scored 130 in text (barely meeting the threshold), its visual score skyrocketed to 145. This 15-point gap is not a testing error. It signifies that the model's ability to "identify patterns in images" exceeds its "text-based reasoning" by a full Standard deviation. According to Mensa's distribution, 130 represents the top 2% of the population, while 145 represents the top 0.1%. In visual tasks, GPT-5.5 Pro is truly one in a thousand.
The disparity between text and visual capabilities is striking. While GPT-5.5 Pro scored 130 in text (barely meeting the threshold), its visual score skyrocketed to 145. This 15-point gap is not a testing error. It signifies that the model's ability to "identify patterns in images" exceeds its "text-based reasoning" by a full Standard deviation. According to Mensa's distribution, 130 represents the top 2% of the population, while 145 represents the top 0.1%. In visual tasks, GPT-5.5 Pro is truly one in a thousand.
The "Little Engine" strategy
Sam Altman used a deceptive Metaphor: the "little engine."
Behind this seemingly soft term lies OpenAI's extremely cold commercial will. Yesterday, OpenAI launched new API products: gpt-5.5 and the Pro version are now available, significantly enhancing Intelligence levels and Token Efficiency, making them perfect for complex task scenarios such as coding and analysis. In the LisanBench test, GPT-5.5 is the strongest non-reasoning dual-sport model.
More chillingly, GPT-5.5 has reduced token usage by 45.6% while increasing intelligence scores by 1.77 times. This means OpenAI is simultaneously "monopolizing intelligence" and "reducing costs through dimensional reduction."
Sam Altman used a deceptive Metaphor: the "little engine."
Behind this seemingly soft term lies OpenAI's extremely cold commercial will. Yesterday, OpenAI launched new API products: gpt-5.5 and the Pro version are now available, significantly enhancing Intelligence levels and Token Efficiency, making them perfect for complex task scenarios such as coding and analysis. In the LisanBench test, GPT-5.5 is the strongest non-reasoning dual-sport model.
More chillingly, GPT-5.5 has reduced token usage by 45.6% while increasing intelligence scores by 1.77 times. This means OpenAI is simultaneously "monopolizing intelligence" and "reducing costs through dimensional reduction."
The Compressed Window
The true weight of this event lies not in the scores themselves, but in the rhythm. Last year, the visual ceiling was 119. This year, it jumped directly to 145. There was no transition, no warning—just a line of updates. In this speed, all "proficiency" is depreciating. As Yuval Noah Harari noted, for the first time in history, we cannot predict what the human job market will look like in 20 years. However, OpenAI may have shortened this window to just 20 months. The "honest little engine" has sounded its whistle; it does not intend to wait for anyone. If you feel turbulence, it is because the era is changing tracks. The half-life of Skills is being repeatedly shortened by this iteration speed—not in ten years, not in five, but by the month.
The true weight of this event lies not in the scores themselves, but in the rhythm. Last year, the visual ceiling was 119. This year, it jumped directly to 145. There was no transition, no warning—just a line of updates. In this speed, all "proficiency" is depreciating. As Yuval Noah Harari noted, for the first time in history, we cannot predict what the human job market will look like in 20 years. However, OpenAI may have shortened this window to just 20 months. The "honest little engine" has sounded its whistle; it does not intend to wait for anyone. If you feel turbulence, it is because the era is changing tracks. The half-life of Skills is being repeatedly shortened by this iteration speed—not in ten years, not in five, but by the month.
Comments & Questions (0)
No comments yet
Be the first to comment!