How close are we to AGI?
AGI (artificial general intelligence), as of October 2026: AI beats PhD experts on hard science questions (96% vs 70%, Gemini 4 Argon) but finishes only 21% of real freelance jobs to a professional's standard (Claude Opus 5.5).
What today's AI still lacks
Memory
Not solved
A regular personRemembers what they learned and did, for years.
Continual learning
Not solved
A regular personGets better at a job every week.
Jagged ability
Not solved
A regular personGood at hard things and easy everyday things alike.
Long tasks alone
Partly solved
A regular personFinishes a whole paid project.
Learning new skills fast
Partly solved
A regular personPicks up a brand-new game in minutes.
Reliability
Partly solved
A regular personDoesn't invent facts.
Invention
Partly solved
A regular personComes up with new ideas in any field.
Own judgment
Partly solved
A regular personDecides what is worth doing.
Where AI already beats people
PhD-level science questions
New picture puzzles
Unsolved maths
Solved Navier-Stokes, a $1 million prize problem · Sep 2026
Sources (14)
- arXiv, Hendrycks et al., A Definition of AGI (arXiv 2510.18212), Mar 2023
- arXiv, arxiv.org/abs/2607.22157
- Scale AI, labs.scale.com/leaderboard/hss
- Scale AI, labs.scale.com/leaderboard/rli
- ARC Prize, OpenAI's GPT-6 Astra on ARC-AGI-3 (ARC Prize blog), with the ARC-AGI-3 launch paper as baseline
- ARC Prize, arcprize.org/arc-agi
- GitHub, github.com/vectara/hallucination-leaderboard
- Epoch AI, epoch.ai/frontiermath/open-problems
- Anthropic, anthropic.com/research/ecdsa-curve-origins
- arXiv, arxiv.org/abs/2610.06824
- Epoch AI, epoch.ai/benchmarks/gpqa-diamond
- ARC Prize, arcprize.org/leaderboard
- ARC Prize, arcprize.org/blog/announcing-arc-agi-2-and-arc-prize-2025
- OpenAI, openai.com/index/navier-stokes-solution