AGI: The race for superintelligence — 2026 could be a turning point

The world is on the verge of a technological leap capable of reshaping the economy, politics, and our understanding of human labor. Some experts predict the emergence of “powerful AI” as soon as 2026, while others caution that artificial general intelligence (AGI) remains a distant peak, comparable in complexity to decoding the human brain. But what exactly is AGI, how close are current models, and is there a chance to see it within a year?

What AGI is and why it’s more than just AI

AGI is defined as a system capable of performing any intellectual task at a human level, transferring knowledge across domains, adapting to new situations, and using common sense. A 2025 MIT Technology Review Insights report emphasizes that AGI involves modeling the brain’s key cognitive functions, not simply increasing algorithmic power.

OpenAI describes AGI as an autonomous system able to outperform humans in economically significant work—from medicine to law. Modern AI models, however impressive in narrow tasks, remain far from true “general” intelligence. They struggle with deep language comprehension, world perception, solving atypical problems, and handling complex context.

Intelligence tests: why AI still doesn’t “understand” the world

François Chollet, creator of the ARC Prize, defines intelligence as the ability to combine known elements into new solutions. His ARC-AGI-2 test measures exactly this flexibility. Here, AI scores 0%, while humans perform almost perfectly.

Research teams note that true intelligence requires balance: logical reasoning, flexibility, social understanding, and the ability to interact with the physical world. Rumman Choudhury of Humane Intelligence suggests considering intelligence more broadly, rather than limiting it to human models.

2026: breakthrough or illusion?

Optimists, including Anthropic co-founder Dario Amodei, believe that by 2026 systems with intelligence comparable to Nobel-level expertise in specific domains may emerge. These systems could switch between text, audio, video, and physical tasks, partially setting their own goals.

OpenAI CEO Sam Altman notes that the term “AGI” is gradually losing meaning, as key properties of such systems are already appearing, and their future impact could be comparable to the arrival of electricity or the internet.

Forecasts from Metaculus and 80,000 Hours estimate a 50% chance of reaching major AGI milestones by 2028, with a 10% probability that machines will surpass humans in all tasks by 2027, rising to 50% by 2047. Timelines are shrinking rapidly—from 50 years in 2020 to roughly five years by late 2024.

Technical barriers: why AGI may be delayed

MIT warns that full AGI will require revolutions in hardware, software, and architectures. Current growth in computational power cannot continue indefinitely, while demand is increasing exponentially.

Development costs may reach levels comparable to the GDP of large nations. Some researchers suggest that transformer architectures are approaching their limits, and true AGI will require fundamentally new designs.

Modern models, despite massive training, can communicate in human language but lack full understanding of the world and continue to fail adaptability tests. This suggests that in 2026, we are more likely to see “proto-AGI”—powerful next-generation systems, but not true general intelligence.