AGI at Scale

We have entered the era of reasoning models and agentic workflows. Here is the factual landscape of Artificial General Intelligence in 2025.

What is AGI?

Artificial General Intelligence represents AI systems with human-like adaptability across all intellectual domains.

Generative AI

  • Excels at specific tasks
  • Requires extensive training
  • Cannot transfer learning
  • Pattern recognition only

Reasoning Agents

  • Adapts to any challenge
  • Learns from minimal examples
  • Transfers knowledge
  • True reasoning

2024–2025 Breakthroughs

The past year has seen unprecedented advances in AI capabilities, pushing us closer to AGI than ever before.

GPT-5.2

Unified Intelligence

100.0%

on AIME 2025 (no tools). OpenAI's most capable model combining reasoning, agents, and multimodal understanding.

Gemini 3

Deep Thinking

1M

token context window. Google's biggest model upgrade yet, with a new Deep Think mode for harder reasoning.

Claude Opus 4.5

Agentic Excellence

80.9%

on SWE-bench Verified. A top-tier model for coding, agents, and computer use.

OpenAI o3

Reasoning Champion

96.7%

on AIME 2024. The successor to o1, with 20% fewer major errors on complex reasoning tasks.

Expert Predictions

When leading researchers and industry leaders expect AGI to arrive.

Swipe to explore
2026
Elon Musk
AI smarter than humans
2027
Dario Amodei
Anthropic CEO
2027
Forecasters
Metaculus Consensus
2029
Sam Altman
OpenAI CEO
~2030
Demis Hassabis
DeepMind CEO
~2040
Researchers
Survey average

Capability Progress

How AI benchmarks have improved from 2024 to 2025.

Competition Math (AIME)12% → 100.0%
PhD-Level Science (GPQA)69.7% (Human) → 92.4% (AI)
Software Engineering (SWE-bench)1.96% → 80.9%
Abstract Reasoning (ARC-AGI-2)0% → 52.9%
Visual Reasoning (MMMU)55% → 84%

Infrastructure at Scale

The massive build-out powering the race to AGI.

460TWh
Global data center electricity (2022)
1,000+TWh
Projected by 2026
200K
GPUs in xAI Colossus (reported)
$500B
Stargate project (commitment)

Key Progress Drivers

The four forces accelerating the path to AGI.

Inference-Time Compute

Models that 'think' for seconds or minutes before answering.

Massive Context

Context windows expanded to 1 Million tokens.

Synthetic Data

Training on high-quality AI-generated reasoning traces.

Price Performance

Claude Haiku 4.5 costs $1/M tokens—100x cheaper than 2023 frontier models.

Remaining Challenges

Technical hurdles still on the path to true AGI.

Episodic Memory

Models still lack long-term, evolvable memory of users.

Reliable Planning

Agents still fail on edge cases during long tasks.

World Modeling

Understanding physics and causality remains difficult.

Energy Constraints

Power grids cannot keep up with GPU cluster demands.

Data Wall

Human-generated training data is nearly exhausted.

Safety & Control

Aligning superhuman reasoning models remains unsolved.

Research Pathways

Societal Integration Technological Advancement Explainability & Trust Ethical Considerations Brain-Inspired Systems

Implications

The opportunities and challenges AGI will bring.

Economic Transformation

AI agents are automating entire workflows, not just tasks.

Scientific Discovery

AlphaFold 3 and AI-driven labs are accelerating research.

Healthcare Revolution

AI assistants are improving diagnostic support and clinical workflows.

Labor Disruption

Knowledge work is being restructured around AI capabilities.

Deepfakes & Truth

Photorealistic AI content challenges information integrity.

Regulation

EU AI Act rollout underway; global governance frameworks emerging.

2025

The window is closing.

Organizations establishing AI readiness now will be positioned to thrive. The key areas: AI strategy, infrastructure, governance, and workforce development.

"AGI, probably the most transformative moment in human history, is on the horizon."

— Demis Hassabis, CEO of Google DeepMind