The Evolutionary Road to AGI & Superintelligence (ASI): Benchmarking 2026 Frontier Models Against Human PhD Ability
From next-token prediction to autonomous theorem proving: tracking ARC-AGI, Humanity’s Last Exam (HLE), and recursive self-improving systems.
How 2026 foundation models crossed from Generative AI into Candidate AGI, and what experimental benchmarks (ARC-AGI-2, HLE, Automated Formal Provers) tell us about the timeline to Artificial Superintelligence (ASI).
1. Crossing the AGI Operational Definition
In late 2026, foundation models like GPT-6.1 Sol Pro and Claude Sonnet 5.5 consistently outscore human PhDs on blind science evaluations (GPQA Diamond >94%), solve real-world enterprise codebase bugs autonomously, and generate formal Lean 4 mathematical proofs. The transition toward Artificial Superintelligence (ASI) is now focused on recursive self-improvement.
Explore Interactive Telemetry
Run real-time spend simulations, head-to-head battle radars, and GPU sizing across 250+ foundation models.
Launch ModelAnalysis Telemetry →