How Close Are We to AGI? Every Major Prediction, Compared (2026)

How close are we to AGI? AGI timeline predictions from lab leaders, Metaculus, superforecasters and researcher surveys, plus what's still missing.

By Izhar Ahmad Danish · · 4 min read

How Close Are We to AGI? Every Major Prediction, Compared (2026): Time of Superintelligence cover image
On this page
  1. AGI timeline: who predicts what
  2. Has AGI already arrived?
  3. How fast are the benchmarks falling?
  4. What’s still missing before AGI?
  5. Why do AGI predictions differ so much?
  6. What happens after AGI?
  7. Our take
  8. Frequently asked questions
  9. Sources

How close are we to AGI? That depends on whom you ask and how they define it. This page gathers the most-cited AGI predictions in one sourced table, shows how the forecasts have moved, and explains what still separates today’s AI from artificial general intelligence. For the definition itself, see what is AGI.

Short answer: As of October 2026, AGI is close but not here by most definitions. AI lab leaders mostly forecast the late 2020s, Metaculus forecasters give 50% odds by 2033, and a large survey of AI researchers said 2047. Benchmarks are nearly saturated; the missing pieces are reliability, long autonomous work and physical-world skills.

AGI timeline: who predicts what

Short answer: industry leaders say 2026–2030, forecasting communities say early 2030s, academic surveys say the 2040s.

Who When said Prediction Type
Jensen Huang, Nvidia Mar 2026 AGI “already achieved” (by his own billion-dollar-company test) Industry
Elon Musk, xAI/SpaceX Jan 2026 (Davos) AI smarter than any single human by end of 2026 or 2027 Industry
Sam Altman, OpenAI Jan 2025 “Confident we know how to build AGI as we have traditionally understood it” Industry
Dario Amodei, Anthropic Oct 2024 Powerful AI “could come as early as 2026” Industry
Demis Hassabis, Google DeepMind 2025 Human-level AI at any task in 5–10 years Industry
AI 2027 (Kokotajlo et al.) 2025 Fast scenario: superhuman AI around 2027 Scenario
Ray Kurzweil 2024 Human-level AI by 2029 Futurist
Samotsvety superforecasters 2023 ~28% chance of AGI by 2030 Forecasters
Metaculus community Feb 2026 25% by 2029; 50% by 2033 Forecasters
2,778 AI researchers (Grace et al.) 2023 survey 50% chance machines beat humans at every task by 2047 Academic
XPT superforecasters 2022 25% chance of AGI by 2048 Forecasters

Sources: each row is linked in the sources list below. Metaculus, Samotsvety and XPT figures as summarized by 80,000 Hours.

Two patterns stand out. First, the closer someone is to building frontier models, the earlier their date. Second, every group has moved earlier. 80,000 Hours notes that researchers in the 2022 survey expected AI to write simple Python code by 2027; a year later they said 2025, and both guesses turned out too pessimistic.

Has AGI already arrived?

Short answer: a few leaders say yes; most researchers say no.

Huang’s March 2026 claim rests on a narrow test. Critics such as Gary Marcus argue that systems which still make basic errors and can’t be trusted to finish open-ended work are not general intelligence. The what is AGI guide covers the definitions in detail; here the practical point is that “AGI” now means different things in different press releases, so always check which definition a claim uses.

How fast are the benchmarks falling?

Short answer: faster than their designers expected.

  • ARC-AGI-2, a puzzle test built to resist memorisation, had GPT-6 at 95% in September 2026 on the ARC Prize leaderboard.
  • On Humanity’s Last Exam, written to favour human experts, frontier models gained 30 percentage points in one year, according to Stanford’s 2026 AI Index.
  • Gemini Deep Think scored 35 of 42 points at the 2025 International Mathematical Olympiad, a gold-medal score.
  • The AI Index says tests meant to last years are now “saturated in months.”

For a full scorecard against human baselines, see is AI smarter than humans?

What’s still missing before AGI?

Short answer: reliability, long-horizon autonomy and the physical world.

The same AI Index that reports record test scores also lists the gaps:

Gap 2026 evidence
Reliable task completion Agents still fail roughly 1 in 3 attempts on structured benchmarks
Using a computer like a person 66.3% on OSWorld vs a 72.35% human baseline
Physical-world tasks Robots succeed in only 12% of real household tasks
Odd blind spots Best model read analog clocks correctly 50.6% of the time vs ~90% for humans

This is why we think “how close are we to AGI?” is now less a question about intelligence tests and more about whether AI can be trusted to do a whole job unsupervised. That is also the bar in OpenAI’s own charter definition: systems that outperform humans at most economically valuable work.

Why do AGI predictions differ so much?

Short answer: different definitions, different incentives, different bets on scaling.

  • Definitions. “Beats humans on exams” arrives years before “does most paid work.”
  • Incentives. Lab leaders raise billions on the promise of AGI. That doesn’t make them wrong, but it is a reason to weigh their dates against independent forecasters.
  • Scaling vs. missing ingredients. Optimists expect more compute and better training to close the remaining gaps. Skeptics such as Yann LeCun argue new ideas, like world models, are needed.

What happens after AGI?

Many forecasters expect superintelligence to follow AGI quickly, because AI that can do AI research could speed up its own progress. We cover that jump in AGI vs ASI, the AI 2027 scenario and our super intelligence timeline.

Our take

The honest range for AGI is “already, by a loose definition” to “2040s, by a strict one.” If you need one number, the Metaculus median of about 2033 is a reasonable midpoint because it pools many forecasters and updates continuously. Watch reliability and agent benchmarks, not IQ-style tests: when AI can finish a week of real work without a human catching its mistakes, the debate will be over. We’ll update this table as new forecasts appear.

Enjoyed this? Get the SI Briefing

One short email a week: what changed on the road to super intelligence, and why it matters. No spam, unsubscribe anytime.

Frequently asked questions

How close are we to AGI?

Closer than most experts expected five years ago, but not there by most definitions. Lab leaders point to the late 2020s, Metaculus forecasters put 50% odds on 2033, and surveyed AI researchers said 2047. The remaining gap is reliability and autonomy in real work, not test scores.

When will AGI arrive?

Nobody knows. The most-cited forecasts range from 'already here' (Jensen Huang, March 2026) through the late 2020s (Altman, Amodei, Musk) and early 2030s (Metaculus) to 2047 (median of a 2023 survey of 2,778 AI researchers).

Is ChatGPT AGI?

Not by most definitions. ChatGPT is general-purpose and scores at or above human level on many tests, but it still fails at tasks a capable person would finish, and it can't reliably do most jobs end to end without supervision.

When will AI be smarter than humans?

On many narrow tests it already is. Being smarter across the board is the AGI question: Elon Musk said at Davos 2026 it could happen by end of 2026 or 2027, while academic surveys put a 50% chance at 2047.

Why do AGI predictions differ so much?

People use different definitions (passing tests vs doing most paid work), have different incentives (lab leaders raise money; academics don't), and weigh scaling trends against unsolved problems such as reliability, memory and physical-world skills differently.

14 Sources

  1. Sam Altman, “Reflections” (January 2025)
  2. Dario Amodei, “Machines of Loving Grace” (October 2024)
  3. NBC News, Hassabis: AI matching humans at any task in five to 10 years (2025)
  4. AI Magazine: Musk on AI and superintelligence at Davos 2026
  5. Fortune, Jensen Huang says ‘we’ve achieved AGI’ (30 Mar 2026)
  6. 80,000 Hours, Shrinking AGI timelines: a review of expert forecasts (Metaculus, Samotsvety, XPT figures)
  7. Grace et al., “Thousands of AI Authors on the Future of AI” (2024)
  8. AI 2027 scenario (AI Futures Project)
  9. The Singularity Is Nearer (Ray Kurzweil, 2024)
  10. ARC Prize, ARC-AGI results
  11. Stanford HAI, 2026 AI Index: Technical performance
  12. OpenAI Charter (definition of AGI)
  13. TechCrunch: Yann LeCun’s AMI Labs raises $1.03B to build world models (Mar 2026)
  14. Gary Marcus, “Sorry, Sam and Elon, we have not reached the Singularity” (28 Jul 2026)
Before you go

Super intelligence is moving fast.

Don’t miss what happens next

Get one short email a week with what changed and why it matters. No spam. Unsubscribe anytime.