The AGI race is on. Track the 2026 milestones, including the Bar Exam, IMO, and the final hurdle: ARC-AGI. Get expert forecasts from the industry's top minds.
Narrow AI (what we have now): A specialist. It can be world-class at chess or lead qualification, but it cannot pivot. A sales bot cannot suddenly decide to write a legal brief or help you debug a quantum physics equation.
General AI (the goal): A generalist. It possesses Transfer Learning—the ability to take a lesson learned in one domain (like logic) and apply it to a completely unrelated one (like music theory) without being specifically programmed to do so.
The Four Pillars of AGI:
Reasoning: The ability to solve Zero-Shot problems (puzzles it has never seen before).
Memory: A persistent, human-like working memory that spans weeks, not just minutes.
Common Sense: An intuitive understanding of the physical world (e.g., "if I drop this glass, it will break").
Autonomy: The ability to set its own sub-goals to achieve a larger objective.
Where Are We Today? (The 2026 Status)
While most experts agree we haven't reached total AGI (the point where a machine can replace a human in every job), we have entered the era of Inaugural AGI or Agentic AGI.
In late 2025, models like OpenAI’s o-series and Google’s Gemini 3 introduced Test-Time Compute. This allows the AI to think for several seconds or minutes before answering, clearing complex benchmarks like the International Math Olympiad and the Bar Exam with ease.
AGI-lite agents are now handling 20-30% of global e-commerce and logistics autonomously, managing entire supply chains without human intervention.
The current holy grail test is ARC-AGI. While humans score ~85% on these novel visual puzzles, the best AI models in early 2026 are still hovering around 45-55%. This is the gap we are currently trying to close
When Do The Experts Think AGI Will Arrive?
The betting window for AGI has moved up significantly since 2024.
Here is the current 2026 AGI Milestone Tracker, showing the human benchmarks AI has officially checked off.
Logic & Math
International Math Olympiad
✅ Cleared
In late 2025, Google and OpenAI models achieved Gold Medal performance (>80% accuracy) on IMO questions.
Professional
Bar Exam & USMLE (Medical)
✅ Cleared
Current frontier models (GPT-5.2, Claude 4.6) now score in the top 10% of human professionals on legal and medical licensing exams.
Creativity
Turing Test (Conversational)
✅ Cleared
In specialized 2025 studies, human judges were unable to distinguish AI from humans in text-based dialogue over 85% of the time.
Social
Theory of Mind (6th Order)
✅ Cleared
Models like GPT-4 and Claude have exceeded adult performance in Theory of Mind —the ability to reason about complex layers of human belief (e.g., "I think that you believe that she knows...").
Engineering
SWE-bench (Software Coding)
🟡 In Progress
AI can now autonomously solve ~72% of complex software engineering tasks that previously required multiple hours of human programmer time.
Fluid Intel.
ARC-AGI (Visual Reasoning)
🔴 Unsolved
While humans score ~85%, the best AI models (GPT-5.2 Pro) currently peak around 43-54% on this test of pure, out-of-the-box reasoning.
❓ Frequently Asked Questions (FAQs)
Why is ARC-AGI the most important test on this list if AI has already passed the Bar Exam?
The Bar Exam and medical tests are essentially tests of memory and pattern matching. Because AI has read almost every book ever written, it can remember the law. ARC-AGI, however, is a test of Fluid Intelligence. It uses simple colored grids with rules the AI has never seen before. To pass it, the AI can't rely on its training; it must learn a new rule in seconds, just like a human child. As of 2026, this is the final boss of AI benchmarks.
What is Chain-of-Thought (CoT) and why did it accelerate these milestones?
Before 2025, AI was reactive—it predicted the next word instantly. In 2026, Reasoning Models (like the o-series or Thinking modes) use Test-Time Compute. This means the AI pauses to think, running internal simulations and checking its own logic before giving you an answer. This internal deliberation is why AI was finally able to jump from a 10% score to an 80% score on the International Math Olympiad.
If AI has passed the Bar Exam, why hasn't it replaced lawyers yet?
Passing a test measures knowledge, but being a lawyer (or a sales rep) requires agency and physical-world interaction. AI still struggles with Long-Horizon Tasks, tasks that take days or weeks of consistent planning, physical movement, and high-stakes social negotiation. In 2026, we are in the era of Inaugural AGI: the digital brain is ready, but the body (autonomous agents and robotics) is still catching up.
If you’ve read this and are excited to see just how AI can help your business reach the next level - come and have a go with our APE AI. Email or chat to our autonomous agents (for free) and if you’ve got any questions, one of our human team will be happy to answer them.
.svg)
