Claude Opus 4.6: The AI That Challenges Human Intelligence

Claude Opus 4.6: A Leap Forward in Artificial Intelligence

Claude Opus 4.6, an advanced artificial intelligence model, boasts an impressive IQ of 155. To put this into perspective, the average IQ of a human is 100, meaning this AI model far exceeds human norms. Even more remarkable is that Claude Opus 4.6 is progressing by 20 IQ points every six months, a development pace that seems straight out of science fiction.

Two Figures That Stand Out

Two figures deserve special attention. While they may seem provocative, they are undeniably true. The first figure pertains to the average IQ of an adult, which is set at 100 according to the Wechsler scale, with a standard deviation of 15. An IQ of 130 is achieved by only 2% of the population, while a score of 145 is extremely rare, approximately 1 in 1,000. An IQ of 160 is almost unheard of, reserved for a few exceptional individuals.

The second figure concerns Claude Opus 4.6, which, in March 2026, is estimated to be 155 on this same scale, placing it in the 99.98th percentile of the human population. This is not mere speculation, but a well-documented development trajectory, with an increase of 20 IQ points every six months.

The Limits of IQ

While IQ is a popular measure of intelligence, it is not without its limitations. It is widely recognized that it does not capture essential aspects of human intelligence such as empathy, pure creativity, leadership, or emotional intelligence. The Wechsler scale, however, focuses on four specific dimensions:

• Verbal comprehension
• Perceptual reasoning
• Working memory
• Processing speed

Comparison on Dimensions of Intelligence

In terms of verbal comprehension, an average human masters one or two languages. In contrast, Claude Opus 4.6 operates in about a hundred languages, without loss of quality, giving the AI an undeniable advantage.

Regarding working memory, an average human can hold about 7 items simultaneously. Claude Opus 4.6, on the other hand, can manage a context window of one million tokens, equivalent to processing 750,000 words in a single pass, without forgetting anything. This represents a ratio of 1 to 100,000.

In terms of processing speed, the human brain processes conscious information at about 20-30 bits per second. Claude Opus 4.6, however, generates about 65 tokens per second, continuously and without fatigue.

Perceptual Reasoning: A Challenge Met

The benchmark ARC-AGI-2, designed by François Chollet, is specifically created to be resistant to memorization. The visual-logical puzzles it contains are novel for models during their training. In 2026, results show that the average human achieves 60% success, while Claude Opus 4.6 reaches 68.8%. Expert humans, serving as a control group, achieve 100%. This is the first time a general artificial intelligence system statistically surpasses the average human on a task of pure abstract reasoning.

Applications in Various Fields

Medicine

In the medical field, Claude Opus 4.6 has demonstrated exceptional capabilities. On the USMLE Step 2 CK, a benchmark medical exam, it achieves a score of 89.3%, placing it beyond the 95th percentile of human candidates. In clinical studies, this model surpasses the diagnostic accuracy of general practitioners and matches that of specialists in 76.1% of complex cases.

Law

In the legal domain, Claude Opus 4.6 has also proven itself. On the BigLaw Bench, which evaluates contract analysis, it scores 90.2%, the highest ever recorded for a model in its category.

Finance

In finance, on the GDPval-AA benchmark that assesses productivity in finance and auditing, Claude Opus 4.6 achieves a score of 1606 Elo, placing it in the top 1% of performances. These results are not mere simulations or prepared demonstrations to impress, but reproduced, verified, and published benchmarks.

The crucial question is no longer whether AI can accomplish what a human expert does. The answer is clearly yes, and this applies to an increasing portion of the spectrum of human tasks.