Frontier AIThe AGI RunwayBrilliant in moments, but not yet dependable over missions. Powerful general-purpose systems for reasoning, coding, creation, analysis, and tool use—still reliant on human direction, verification, and judgment.
TRACKING THE FRONTIER. MAPPING THE SINGULARITY.
The Path to General ASI
ASIndex independently measures the demonstrated general capability of broadly available AI systems—and maps the specialized superintelligences and recursive improvement loops that may accelerate the Singularity.
Current leader: GPT-6 ASTRA by OpenAI Current phase: AGI-I — Human-Level General Intelligence
24.9 / 100
ASIndex General Capability Score
AGI-IHuman-Level General IntelligenceThe machine crosses from assistant to cognitive peer. It adapts to unfamiliar work, plans and executes multi-step tasks, uses tools, and delivers broadly competent results with limited oversight.
AGI-IIExpert General IntelligenceExpertise becomes general. The system matches strong human specialists across many domains, exceeds them in some, and transfers insight between fields without losing rigor.
AGI-IIIExceptional General IntelligenceIt does not just navigate the frontier—it moves it. Better than nearly every individual human across most cognitive work, it repeatedly produces original discoveries, designs, and strategies.
General ASI RunwaySuperhuman General IntelligenceNo individual human remains the benchmark. The system exceeds the best human experts across virtually all cognitive domains and sustains complex, long-horizon programs with extraordinary reliability.
General ASICivilization-Scale General SuperintelligenceNo human institution remains the benchmark. General ASI outperforms even large, well-coordinated human expert collectives across virtually all cognitive work, operating at machine speed and scale.100 marks ASIndex’s General ASI threshold—not the ceiling.
Scores represent demonstrated general capability—not elapsed time, percentage completion, probability, or a forecast of when the next threshold will be reached.
This score measures broad, publicly available AI capability. A system may achieve specialized superhuman—or specialized superintelligent—performance within a bounded domain without scoring highly on this general-capability scale.
The destination is not intelligence alone. It is aligned superintelligence that expands human agency, accelerates discovery, and helps civilization thrive.
Top 5 Leaderboard
Anti-Benchmaxxing Methods AppliedAmong broadly available AI models, these are the farthest along the path to ASI.
GPT-6 ASTRAMAX
OpenAI
Cost$$$
Speed⚡⚡
24.9
FABLE 5.1MAX
Anthropic
Cost$$$$
Speed⚡⚡
24.1
MUSE SPARKMAX
Meta
Cost$$
Speed⚡⚡⚡⚡⚡
19.8
GROK 4.6XHIGH
SpaceXAI
Cost$$
Speed⚡⚡⚡
19.6
KIMI K3MAX
Moonshot AI
Cost$$
Speed⚡⚡
18.8
Scores are point-in-time estimates of publicly available AI system capability. Hosted systems can improve over time even when the public name stays the same, so we will test weekly with fresh test mechanics and algorithms.
THE ASINDEX SINGULARITY THESIS
The Singularity is not a moment, it is a Journey
Specialized systems may become superintelligent within valuable domains before broadly capable AGI arrives. AGI may then provide the orchestration layer: understanding objectives, decomposing missions, selecting specialists, reconciling evidence, managing uncertainty, and preserving long-term direction.
The specialists and general orchestrator may improve one another through ecosystem-level RSI, accelerating progress toward General ASI.
Specialized ASI→AGI Orchestrator⇄S-ASI Network→General ASI
Specialized ASI with Recursive Self-Improvement (RSI)
Our Singularity ThesisASIndex measures public AI system capability, not safety certification, investment advice, procurement advice, or regulatory approval.