The AGI Timeline Tracker puts 32 published forecasts for when AGI arrives on one timeline: the AI 2027 scenario and its authors' later updates, Metaculus, expert surveys and AI lab leaders, each with a date and source. Pick a milestone, set your own year, and see where you sit against the crowd.
Pick a milestone, see every tracked forecast on one timeline, then set your own year. Nothing leaves your browser.
AGI: AI that matches or beats humans at most cognitive or economically valuable work. Sources call it AGI, human-level AI or "powerful AI".
You vs 16 forecasts
Close to the middle of the pack
6 expect it later than 2029, 8 expect it sooner, 2 say 2029. Median of tracked forecasts: 2029.
Elon Musk · 2026 (2026-2027)
CEO, xAI / Tesla · said Jan 2026
"I think we might have AI that is smarter than any human by the end of this year... No later than next year." Source for Elon Musk (opens in a new tab)Davos 2026, with Larry Fink. Smarter than any single human.
Jensen Huang · 2026
CEO, NVIDIA · said Mar 2026
"I think we've achieved AGI." Source for Jensen Huang (opens in a new tab)Says AGI is already here under a narrow, commercial definition: AI that could build a billion-dollar business (Lex Fridman podcast).
Mustafa Suleyman · 2027
CEO, Microsoft AI · said Feb 2026
On white-collar computer work: "most of those tasks will be fully automated by an AI within the next 12 to 18 months." Source for Mustafa Suleyman (opens in a new tab)Converted from 12 to 18 months (Financial Times interview). About automating most white-collar computer tasks, not AGI as such.
Dario Amodei · 2027 (2027-2028)
CEO, Anthropic · said Feb 2026
"I really do believe that we could have models that are a country of geniuses in the data center in one to two years." Source for Dario Amodei (opens in a new tab)"Country of geniuses in a datacenter", converted from "one to two years" (Dwarkesh Podcast). Anthropic's Mar 2025 policy submission said late 2026 or early 2027.
Daniel Kokotajlo · 2027
Executive Director, AI Futures Project (AI 2027 lead author) · said Nov 2026
Top-expert-dominating AI median "Sep 2027 (from late 2028)". Source for Daniel Kokotajlo (opens in a new tab)TED-AI: AI that beats top human experts at nearly all cognitive tasks, a stronger bar than most AGI definitions.
Metaculus: weakly general AI · 2027 (2026-2029)
Community median, 1,730 forecasters (question 3479) · said Nov 2026
Community median about Aug 18, 2027; middle half of forecasts Sep 2026 to Sep 2029. Source for Metaculus: weakly general AI (opens in a new tab)Read Nov 12, 2026 from the public embed. Was about Sep 2028 in early Sep 2026.
Shane Legg · 2028
Co-founder & Chief AGI Scientist, Google DeepMind · said Dec 2025
50% chance of "minimal AGI" by 2028; full AGI three to six years later. Source for Shane Legg (opens in a new tab)A forecast he has held since 2009 (interview with Hannah Fry).
Eli Lifland · 2028
AI Futures Project forecaster (AI 2027 co-author) · said Nov 2026
"I'm at a SAR median of August 2027, and a TED-AI median of September 2028." Source for Eli Lifland (opens in a new tab)TED-AI (top-expert-dominating AI) median Sep 2028, previously mid 2032.
Ray Kurzweil · 2029
Futurist; Principal Researcher, Google · said Jun 2024
"So 2029, both for human-level intelligence and for artificial general intelligence (AGI)" Source for Ray Kurzweil (opens in a new tab)A date he has held since 1999. Interview with The Observer.
Demis Hassabis · 2029 (2028-2029)
CEO, Google DeepMind · said Oct 2026
WSJ Magazine profile: Hassabis now expects AGI "within two to three years"; "It's soon... I'm updating nearer and nearer." Source for Demis Hassabis (opens in a new tab)Converted from "two to three years" (WSJ Magazine profile, via Tech Times). Earlier in 2026 he said 5 to 10 years, then 2030.
Metaculus: general AI · 2030 (2028-2036)
Community median, 1,848 forecasters (question 5121) · said Nov 2026
Community median about Jun 30, 2030; middle half of forecasts Feb 2028 to Jun 2036. Source for Metaculus: general AI (opens in a new tab)Strong definition that includes robotics. Read Nov 12, 2026 from the public embed; it was about Jan 2033 in Sep 2026.
Andrej Karpathy · 2035
Founder, Eureka Labs (ex-OpenAI, ex-Tesla AI) · said Oct 2025
"It will take about a decade to work through all of those issues." Source for Andrej Karpathy (opens in a new tab)Converted from "about a decade" (Dwarkesh Podcast episode "AGI is still a decade away"). About AI agents that really work.
2,778 AI researchers (2023 survey) · 2047
AI Impacts Expert Survey on Progress in AI, Grace et al. · said Jan 2024
"the chance of unaided machines outperforming humans in every possible task was estimated at 10% by 2027, and 50% by 2047." Source for 2,778 AI researchers (2023 survey) (opens in a new tab)High-level machine intelligence. Survey run Oct 2023, published Jan 2024.
Superforecasters (FRI LEAP panel) · 2047
Forecasting Research Institute, Wave 8 · said Jun 2026
"Superforecasters give a similar median probability (80%) and a slightly earlier median year of occurrence (2047)." Source for Superforecasters (FRI LEAP panel) (opens in a new tab)AGI: better than 90% of professionals on 90% of non-physical tasks at no more than 5x human cost.
Experts (FRI LEAP panel) · 2050
Forecasting Research Institute, Wave 8: AI scientists, industry, policy researchers, economists · said Jun 2026
"experts give a median forecast of this occurring in 2050." Source for Experts (FRI LEAP panel) (opens in a new tab)Same AGI definition as the superforecasters; median year assumes it happens before 2100.
2,778 AI researchers (2023 survey) · 2116
AI Impacts Expert Survey on Progress in AI, Grace et al. · said Jan 2024
"all human occupations becoming fully automatable was forecast to reach 10% by 2037, and 50% as late as 2116" Source for 2,778 AI researchers (2023 survey) (opens in a new tab)Full automation of all occupations, a much higher bar than AGI. Shown at the right edge of the chart.
The dark vertical line is your year. Dots after 2060 sit on the right edge. Data checked November 13, 2026.
What the AI 2027 scenario described for each period, next to real, dated events.
Mid 2025 · scenario
Stumbling Agents
"The world sees its first glimpse of AI agents." Computer-using agents are marketed as personal assistants but are unreliable and expensive; coding and research agents start changing some professions.
Late 2025 · scenario
The World's Most Expensive AI
"OpenBrain is building the biggest datacenters the world has ever seen," aiming at a model trained with 10^28 FLOP (about 1,000x GPT-4) and at AI that speeds up AI research; Agent-1 is trained.
Early 2026 · scenario
Coding Automation
"The bet of using AI to speed up AI research is starting to pay off." OpenBrain makes algorithmic progress 50% faster than it would without AI assistants; Agent-1 is released publicly.
Mid 2026 · scenario
China Wakes Up
"In China, the CCP is starting to feel the AGI." Compute-constrained China centralizes research under DeepCent and plans to steal OpenBrain's weights.
Late 2026 · scenario
AI Takes Some Jobs
OpenBrain releases Agent-1-mini, 10x cheaper than Agent-1; the stock market is up 30% in 2026; the junior software engineer job market is in turmoil; the DoD expands contracts with OpenBrain.
2025-05-22 · real
Anthropic launches Claude Opus 4 and Claude Sonnet 4, its first Claude 4 models, with extended thinking.
2025-07-21 · real
An advanced Gemini Deep Think officially reaches gold-medal standard at the International Mathematical Olympiad (35/42); OpenAI reports a gold-level result the same week.
2025-08-07 · real
OpenAI releases GPT-5 (74.9% SWE-bench Verified, 94.6% AIME 2025 without tools).
2025-09-23 · real
OpenAI, Oracle and SoftBank announce five new Stargate datacenter sites, bringing planned capacity to nearly 7 GW and >$400B over three years.
2025-11-18 · real
Google releases Gemini 3 Pro: first on LMArena (1501 Elo), 37.5% on Humanity's Last Exam without tools.
2025-11-24 · real
Anthropic releases Claude Opus 4.5; it scored higher than any human candidate on Anthropic's 2-hour performance-engineering take-home exam.
2025-12 · real
METR measures Claude Opus 4.5's 50% time horizon at about 4 h 49 min (95% CI 1h49m-20h25m), the longest it had recorded; 80% horizon ~27 min.
2026-07-24 · real
Anthropic releases Claude Opus 5.
2026-09-03 · real
OpenAI begins a phased rollout of GPT-6 Astra, the first model to reach its internal 'Critical' cybersecurity threshold; cyber partners get access first.
2026-09-22 · real
METR's pre-deployment evaluation: Opus 5.5 is 'unlikely to be able to fully automate AI R&D' but likely to noticeably accelerate researchers.
2026-10-13 · real
Google rolls out Gemini 4 Argon globally, including in the Gemini app for AI Pro and Ultra subscribers.
2026-11-05 · real
METR: AI R&D at Anthropic ~2x faster
METR's independent assessment finds Anthropic is likely improving its models at about twice the historical rate, with ~80% of the acceleration likely due to AI (substantial uncertainty; much evidence redacted).
Choose superhuman coder, automated AI research, AGI or superintelligence.
Each row shows who made the forecast, when, the year, a quote and a link to the source.
Drag the slider to your own prediction and see how many forecasts expect it sooner or later.
Copy the link or the text of your full timeline. Your years are saved in the link.
Nobody knows, and forecasts run from "already here" to after 2100. The AGI forecasts tracked here have a median of about 2029. Metaculus forecasters expect general AI around mid-2030, AI lab leaders such as Dario Amodei and Demis Hassabis say 2027 to 2029, and the 2023 survey of 2,778 AI researchers gave a 50% chance of human-level machine intelligence by 2047.
AI 2027 is a scenario published in April 2025 by Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland and Romeo Dean of the AI Futures Project. It describes a superhuman AI coder in March 2027, a superhuman AI researcher in August 2027 and superintelligence by December 2027. The authors called 2027 their most likely year, but their median forecasts were later.
Yes, several times. Their medians moved later through 2025: Kokotajlo went from 2028 to 2029 for a superhuman coder, and their December 2025 model said December 2031. They moved earlier through 2026. On November 10, 2026 all six AI Futures Project forecasters said a superhuman AI researcher will likely arrive in 2027 (Kokotajlo May 2027, Lifland August 2027), mainly because METR found AI is speeding up AI research at Anthropic.
On November 12, 2026 the Metaculus community median for the first general AI system (question 5121, which includes robotics) was about June 2030, down from about January 2033 in September 2026. For weakly general AI (question 3479) it was about August 2027.
They use different definitions, from an AI that codes as well as top engineers to the full automation of every job, and different methods: trend extrapolation, surveys and insider judgment. AI lab leaders tend to give the earliest dates and broad surveys of researchers the latest. Check each forecast's definition before comparing years.
Forecasts were checked against their sources on November 13, 2026. Where a person said "in N years", the year is counted from when they said it. Forecasts are opinions about an uncertain future, not facts; AI Wins is not affiliated with any person or organization listed.
See how exposed your job is to AI automation.
Open toolPlain-English definitions of AGI, LLMs, agents and more.
Open toolIs AI a bubble? Score 10 warning signs vs the dot-com era.
Open toolCompare the AI assistants you can use right now.
Open toolTimelines move with every model release. Get the AI news that matters in the free daily AI Wins digest.
Get the free digestOnly positive AI stories, delivered daily or weekly. Free, no spam, unsubscribe anytime.