AGI Research Overview
Tracking progress toward Artificial General Intelligence
Artificial General Intelligence (AGI) represents AI systems capable of matching or exceeding human-level cognitive abilities across virtually all domains. Here we track the leading organizations, their approaches, and timeline predictions from industry experts.
Leading AGI Research Organizations
OpenAI
San Francisco, California
2015
~2035 ("few thousand days")
GPT-5.4 achieves 83% human-level on knowledge work; native computer control
Anthropic
San Francisco, California
2021
2027
Safety-first development with Constitutional AI
Google DeepMind
London, United Kingdom
2010
50% by 2028 (updated)
Physical AI: Gemini powering Atlas robots at Hyundai factories
Meta AI
Menlo Park, California
2013
Long-term (world models needed)
Open-source leadership to make Llama industry standard
Microsoft Research
Redmond, Washington
1991
2026-2028 (MIT report)
Humanist superintelligence that serves humanity
xAI
San Francisco, California
2023
2026
Massive compute infrastructure; Grok now powers Optimus robots
Cohere
Toronto, Canada
2019
Not pursuing AGI
Enterprise-first, not chasing AGI
Inflection AI
Palo Alto, California
2022
Pivoted to enterprise
Enterprise API-first after Microsoft acquisition
Expert Timeline Predictions
AGI Timeline Predictions by Expert
| Expert | Affiliation | Date | Prediction |
|---|---|---|---|
| Elon Musk | xAI, Tesla | Dec 2025 | 2026 |
| Shane Legg | DeepMind | Feb 2026 | 50% by 2028 |
| Dario Amodei | Anthropic | Jan 2026 | 2027 |
| Masayoshi Son | SoftBank | Feb 2025 | 2027-2028 |
| Metaculus Forecast | Community | Feb 2026 | 50% by 2033 |
| Jensen Huang | NVIDIA | Mar 2024 | 2029 |
| Demis Hassabis | DeepMind | Jan 2026 | 50% by 2030 |
| Eric Schmidt | Former Google CEO | Apr 2025 | 2028-2030 |
| Ray Kurzweil | Futurist | 2024 | 2032 |
| Sam Altman | OpenAI | 2024 | ~2035 |
| AI Researchers Survey | Academia | 2023 | 50% by 2047 |
Technical Approaches to AGI
Large Language Models
Transformer-based models with native computer-use and 1M+ token context
Vision-Language-Action (VLA)
End-to-end neural networks that translate visual inputs to motor actions
World Models
Internal representations for simulating physical world
Neuro-Symbolic AI
Hybrid neural networks with symbolic reasoning
Agentic AI
Autonomous action, planning, and task execution across multi-week horizons
Physical AI / Embodied Intelligence
AI systems designed to operate robots in the real world
Important Note
Timeline predictions for AGI vary significantly among experts and are subject to considerable uncertainty. These predictions should be viewed as informed opinions rather than definitive forecasts. The path to AGI involves numerous technical, ethical, and safety challenges that remain unresolved.
Learn about AI Safety & Ethics