{"asOf":"2026-10-11","openedAt":"2026-10-11","live":false,"updatedAt":null,"definition":{"title":"What Laurenceo means by AGI","text":"AI that can do essentially all of the cognitive work a skilled remote professional does, including AI research itself, at comparable cost. Because no single test settles that, the observatory tracks the milestone leading forecasters use as its operational marker: AI that can fully automate coding and AI research (the AI Futures Project's Automated Coder), together with community forecasts of general AI and measured capability trends.","caveat":"Different sources define the milestone differently. Each source is shown with its own definition, and the estimate pools only sources whose definition is close to this one."},"estimate":{"asOf":"2026-10-11","central":"2028-06","low":"2027-11","high":"2029-04","n":4,"confidence":"medium","centralYear":2028.46,"lowYear":2027.85,"highYear":2029.29,"centralLabel":"Jun 2028","lowLabel":"Nov 2027","highLabel":"Apr 2029","line":"Combining the leading published forecasts, AI that can do the work of AI researchers is most likely to arrive around Jun 2028, with the middle of the forecasts between Nov 2027 and Apr 2029.","company":{"id":"openai","name":"OpenAI","margin":"narrow","current":2,"runnerUp":"Anthropic"},"country":{"id":"US","name":"United States","margin":"clear"},"today":{"material":true,"text":"Observatory opened: estimate Jun 2028, likely range Nov 2027 to Apr 2029, from 4 counted sources."},"away":"about 1 year 8 months from now","moved":{"months":-30,"text":"2 years 6 months earlier than on 2025-12-31","span":"2 years 6 months","from":"2030-12","fromLabel":"Dec 2030","fromDate":"2025-12-31"},"href":"/intelligence/forecasts#agi","widthYears":1.44,"windowRule":"middle half of the forecasts"},"sources":[{"id":"aifp-kokotajlo","name":"AI Futures Project: Daniel Kokotajlo","kind":"expert-model","metric":"Automated Coder, median arrival","definition":"AI that can fully automate coding at a leading AI company","url":"https://blog.aifutures.org/","pooled":true,"stale":false,"median":"2027-11","probability":null,"year":2027.871,"date":"2026-08-16","stated":"November 2027","note":"Q2.5 update; timelines got slightly shorter.","sourceUrl":"https://blog.aifutures.org/feed","previous":{"date":"2026-04-02","median":"2028-06","probability":null},"shiftMonths":-7,"observations":3,"history":[{"date":"2025-12-31","median":"2029-10"},{"date":"2026-04-02","median":"2028-06"},{"date":"2026-08-16","median":"2027-11"}],"kindLabel":"Expert model","label":"Nov 2027","whyNot":null},{"id":"aifp-lifland","name":"AI Futures Project: Eli Lifland","kind":"expert-model","metric":"Automated Coder, median arrival","definition":"AI that can fully automate coding at a leading AI company","url":"https://blog.aifutures.org/","pooled":true,"stale":false,"median":"2030-01","probability":null,"year":2030.04,"date":"2026-08-16","stated":"January 2030","note":"Q2.5 update.","sourceUrl":"https://blog.aifutures.org/feed","previous":{"date":"2026-04-02","median":"2030-06","probability":null},"shiftMonths":-5,"observations":3,"history":[{"date":"2025-12-31","median":"2032-02"},{"date":"2026-04-02","median":"2030-06"},{"date":"2026-08-16","median":"2030-01"}],"kindLabel":"Expert model","label":"Jan 2030","whyNot":null},{"id":"aifp-brendan","name":"AI Futures Project: Brendan Halstead","kind":"expert-model","metric":"Automated Coder, median arrival","definition":"AI that can fully automate coding at a leading AI company","url":"https://blog.aifutures.org/","pooled":true,"stale":false,"median":"2029-01","probability":null,"year":2029.04,"date":"2026-08-16","stated":"January 2029","note":"Q2.5 update.","sourceUrl":"https://blog.aifutures.org/feed","previous":null,"shiftMonths":null,"observations":1,"history":[{"date":"2026-08-16","median":"2029-01"}],"kindLabel":"Expert model","label":"Jan 2029","whyNot":null},{"id":"espai-2023","name":"AI Impacts expert survey (2023)","kind":"survey","metric":"High-level machine intelligence, 50% chance","definition":"Unaided machines can accomplish every task better and more cheaply than human workers","url":"https://arxiv.org/abs/2401.02843","pooled":false,"stale":true,"median":"2047-01","probability":null,"year":2047.039,"date":"2024-01-05","stated":"2047","note":"Survey of 2,778 published AI researchers, run in late 2023.","sourceUrl":"https://arxiv.org/abs/2401.02843","previous":null,"shiftMonths":null,"observations":1,"history":[],"kindLabel":"Researcher survey","label":"Jan 2047","whyNot":"older than 24 months"},{"id":"openai-target","name":"OpenAI stated target","kind":"lab-target","metric":"Automated AI researcher","definition":"A system that can carry out AI research projects largely on its own","url":"https://openai.com/index/research-acceleration-view-inside-openai/","pooled":false,"stale":false,"median":"2028-03","probability":null,"year":2028.203,"date":"2026-09-10","stated":"March 2028","note":"Restated after OpenAI said it reached its September 2026 automated research intern goal. A company goal, not a forecast.","sourceUrl":"https://techradar.com/pro/openai-says-it-has-built-an-automated-research-intern-to-carry-out-menial-tasks-and-its-only-just-getting-started","previous":null,"shiftMonths":null,"observations":1,"history":[],"kindLabel":"Company target","label":"Mar 2028","whyNot":"a company goal, not a forecast"},{"id":"polymarket-openai-agi-2026","name":"Polymarket","kind":"market","metric":"OpenAI announces it has achieved AGI before 2027","definition":"OpenAI officially announces AGI","url":"https://www.orrery.me/markets/openai-announces-it-has-achieved-agi-before-2027","pooled":false,"stale":false,"median":null,"probability":14,"year":null,"date":"2026-05-18","stated":"14%","note":"Thin liquidity (about $9.1K visible). A market price, not evidence.","sourceUrl":"https://www.orrery.me/markets/openai-announces-it-has-achieved-agi-before-2027","previous":null,"shiftMonths":null,"observations":1,"history":[],"kindLabel":"Prediction market","label":"14%","whyNot":"a bet on an announcement, not on capability"},{"id":"trend","name":"AI task-length trend (METR data)","kind":"trend","metric":"METR 50% time horizon reaches one work-month (167 hours)","definition":"Extends the record METR time horizon at its fitted doubling time, and dates the month it reaches the threshold with at least even odds.","url":null,"pooled":true,"stale":false,"median":"2027-10","probability":null,"year":2027.786,"date":"2026-10-11","stated":"Oct 2027","note":"Fitted on 7 METR points; the record doubles about every 4 months.","sourceUrl":null,"previous":null,"shiftMonths":null,"observations":7,"history":[],"kindLabel":"Measured trend","label":"Oct 2027","whyNot":null}],"trend":{"id":"trend","name":"AI task-length trend (METR data)","kind":"trend","metric":"METR 50% time horizon reaches one work-month (167 hours)","definition":"Extends the record METR time horizon at its fitted doubling time, and dates the month it reaches the threshold with at least even odds.","url":null,"pooled":true,"stale":false,"median":"2027-10","probability":null,"year":2027.786,"date":"2026-10-11","stated":"Oct 2027","note":"Fitted on 7 METR points; the record doubles about every 4 months.","sourceUrl":null,"previous":null,"shiftMonths":null,"observations":7,"history":[],"kindLabel":"Measured trend","label":"Oct 2027","whyNot":null},"history":[{"date":"2025-12-31","central":"2030-12","low":"2029-07","high":"2035-11","n":4,"stored":false},{"date":"2026-01-29","central":"2029-10","low":"2028-04","high":"2032-02","n":3,"stored":false},{"date":"2026-03-20","central":"2029-10","low":"2027-09","high":"2032-02","n":3,"stored":false},{"date":"2026-04-02","central":"2028-06","low":"2027-09","high":"2030-06","n":3,"stored":false},{"date":"2026-06-26","central":"2028-06","low":"2027-10","high":"2030-06","n":3,"stored":false},{"date":"2026-08-16","central":"2028-06","low":"2027-11","high":"2029-04","n":4,"stored":false},{"date":"2026-10-11","central":"2028-06","low":"2027-11","high":"2029-04","n":4,"stored":true}],"company":{"id":"openai","name":"OpenAI","country":"US","current":2,"recent":10,"measures":["ARC-AGI-2","ARC-AGI-3"],"margin":"narrow","runnerUp":{"id":"anthropic","name":"Anthropic","current":2,"recent":7}},"country":{"id":"US","name":"United States","current":4,"recent":17,"margin":"clear"},"labs":[{"id":"openai","current":2,"recent":10,"total":12,"last":"2026-09-03","name":"OpenAI","country":"US","measures":["ARC-AGI-2","ARC-AGI-3"]},{"id":"anthropic","current":2,"recent":7,"total":8,"last":"2026-09-22","name":"Anthropic","country":"US","measures":["METR time horizon","Epoch Capabilities Index"]}],"countries":[{"id":"US","current":4,"recent":17,"total":20,"last":"2026-09-22","name":"United States"}],"measures":[{"id":"metr-50-time-horizon-hours","name":"METR 50% time horizon (software/ML/cyber tasks)","short":"METR time horizon","unit":"hours (human task duration at which the agent succeeds 50% of the time)","records12":3,"record":{"value":12,"label":"Claude Opus 4.6 (current logistic model)","lab":"anthropic","labName":"Anthropic","date":"2026-03-20","sourceUrl":"https://metr.org/notes/2026-03-20-impact-of-modelling-assumptions-on-time-horizon-results/"},"us":{"value":12,"label":"Claude Opus 4.6 (current logistic model)","lab":"anthropic","date":"2026-03-20","labName":"Anthropic"},"cn":null},{"id":"arc-agi-2-best-verified-pct","name":"ARC-AGI-2 best verified score (semi-private)","short":"ARC-AGI-2","unit":"percent","records12":3,"record":{"value":95,"label":"GPT-6 Astra (top reasoning level)","lab":"openai","labName":"OpenAI","date":"2026-09-03","sourceUrl":"https://arcprize.org/results/openai-gpt-6-astra"},"us":{"value":95,"label":"GPT-6 Astra (top reasoning level)","lab":"openai","date":"2026-09-03","labName":"OpenAI"},"cn":{"value":61.3,"label":"DeepSeek V4 Pro 0813 (max effort)","lab":"deepseek","date":"2026-08-13","labName":"DeepSeek"}},{"id":"arc-agi-3-best-verified-pct","name":"ARC-AGI-3 best reported score","short":"ARC-AGI-3","unit":"percent","records12":4,"record":{"value":62.7,"label":"GPT-6 Astra, standard harness (max reasoning, $26,098)","lab":"openai","labName":"OpenAI","date":"2026-09-03","sourceUrl":"https://arcprize.org/blog/astra"},"us":{"value":62.7,"label":"GPT-6 Astra, standard harness (max reasoning, $26,098)","lab":"openai","date":"2026-09-03","labName":"OpenAI"},"cn":null},{"id":"epoch-capabilities-index","name":"Epoch Capabilities Index (ECI), top model at each date","short":"Epoch Capabilities Index","unit":"ECI points (scale set so Claude 3.5 Sonnet = 130, GPT-5 = 150)","records12":8,"record":{"value":167.35,"label":"Claude Opus 5.5","lab":"anthropic","labName":"Anthropic","date":"2026-09-22","sourceUrl":"https://epoch.ai/data/eci_scores.csv"},"us":{"value":167.35,"label":"Claude Opus 5.5","lab":"anthropic","date":"2026-09-22","labName":"Anthropic"},"cn":null}],"board":[{"id":"openai-announces-agi-2026","question":"Will OpenAI officially announce that it has achieved AGI by 2026-12-31?","probability":3,"deadline":"2026-12-31","category":"Recursive AI R&D","likelihood":"Very unlikely"},{"id":"ai-futures-daniel-2027","question":"Will the AI Futures Project publish an update with an Automated Coder median of 2027.0 or earlier for Daniel Kokotajlo by 2027-06-30?","probability":27,"deadline":"2027-06-30","category":"Recursive AI R&D","likelihood":"Unlikely"},{"id":"metr-2x-acceleration","question":"Will METR publish an evaluation concluding that a frontier model gives at least 2x overall acceleration of AI capabilities progress by 2027-09-30?","probability":33,"deadline":"2027-09-30","category":"Recursive AI R&D","likelihood":"Unlikely"},{"id":"lab-ai-rd-threshold-crossed","question":"Will OpenAI or Anthropic state that a model has crossed its AI self-improvement or AI R&D automation threshold by 2027-12-31?","probability":45,"deadline":"2027-12-31","category":"Recursive AI R&D","likelihood":"Toss-up"},{"id":"openai-automated-ai-researcher-2028","question":"Will OpenAI state that it has an automated AI researcher, with independent evaluator confirmation, by 2028-04-30?","probability":17,"deadline":"2028-04-30","category":"Recursive AI R&D","likelihood":"Unlikely"},{"id":"metr-time-horizon-24h","question":"Will METR publish a 50% time horizon point estimate of at least 24 hours for any AI model by 2027-06-30?","probability":58,"deadline":"2027-06-30","category":"Capability","likelihood":"Toss-up"},{"id":"arc-agi-3-90-standard-harness","question":"Will the ARC Prize leaderboard show a verified ARC-AGI-3 score of at least 90% under its standard (non Provider Adapter) harness by 2027-03-31?","probability":55,"deadline":"2027-03-31","category":"Capability","likelihood":"Toss-up"},{"id":"epoch-eci-175","question":"Will any model score at least 175 on Epoch AI's Capabilities Index (ECI) by 2027-09-30?","probability":78,"deadline":"2027-09-30","category":"Capability","likelihood":"Likely"},{"id":"chinese-open-weight-arc-agi-2-85","question":"Will an open-weight model from a Chinese lab reach a verified ARC-AGI-2 semi-private score of at least 85% by 2027-06-30?","probability":62,"deadline":"2027-06-30","category":"Frontier models","likelihood":"Toss-up"},{"id":"arc-agi-3-best-verified-pct-80-2028-04","question":"Will a verified ARC-AGI-3 score of 80% or more be reported by April 30, 2028?","probability":51.9,"deadline":"2028-04-30","category":"Capability","likelihood":"Toss-up"}],"timeline":[{"date":"2026-09-22","kind":"milestone","lab":"anthropic","labName":"Anthropic","title":"METR: about 1.5x AI-driven acceleration","text":"METR's evaluation of Claude Opus 5.5 preliminarily estimates 1.5 years of progress per year from AI, and finds it unlikely to fully automate AI R&D.","url":"https://metr.org/blog/2026-09-22-claude-opus-5-5/"},{"date":"2026-09-22","kind":"record","lab":"anthropic","labName":"Anthropic","title":"Epoch Capabilities Index record: 167.35","text":"Claude Opus 5.5","url":"https://epoch.ai/data/eci_scores.csv"},{"date":"2026-09-10","kind":"forecast","lab":null,"labName":null,"title":"OpenAI stated target: Mar 2028","text":"Restated after OpenAI said it reached its September 2026 automated research intern goal. A company goal, not a forecast.","url":"https://techradar.com/pro/openai-says-it-has-built-an-automated-research-intern-to-carry-out-menial-tasks-and-its-only-just-getting-started"},{"date":"2026-09-06","kind":"milestone","lab":"openai","labName":"OpenAI","title":"OpenAI reports an automated research intern","text":"A system that carries out well-defined research tasks under human direction; no independent verification yet.","url":"https://openai.com/index/research-acceleration-view-inside-openai/"},{"date":"2026-09-03","kind":"record","lab":"openai","labName":"OpenAI","title":"ARC-AGI-2 record: 95%","text":"GPT-6 Astra (top reasoning level)","url":"https://arcprize.org/results/openai-gpt-6-astra"},{"date":"2026-09-03","kind":"record","lab":"openai","labName":"OpenAI","title":"ARC-AGI-3 record: 62.7%","text":"GPT-6 Astra, standard harness (max reasoning, $26,098)","url":"https://arcprize.org/blog/astra"},{"date":"2026-09-03","kind":"record","lab":"openai","labName":"OpenAI","title":"Epoch Capabilities Index record: 166.51","text":"GPT-6 Astra","url":"https://epoch.ai/data/eci_scores.csv"},{"date":"2026-09-01","kind":"record","lab":"anthropic","labName":"Anthropic","title":"Epoch Capabilities Index record: 164.82","text":"Claude Fable 5.1","url":"https://epoch.ai/data/eci_scores.csv"},{"date":"2026-08-16","kind":"forecast","lab":null,"labName":null,"title":"AI Futures Project: Daniel Kokotajlo: Nov 2027","text":"Q2.5 update; timelines got slightly shorter.","url":"https://blog.aifutures.org/feed"},{"date":"2026-08-16","kind":"forecast","lab":null,"labName":null,"title":"AI Futures Project: Eli Lifland: Jan 2030","text":"Q2.5 update.","url":"https://blog.aifutures.org/feed"},{"date":"2026-08-16","kind":"forecast","lab":null,"labName":null,"title":"AI Futures Project: Brendan Halstead: Jan 2029","text":"Q2.5 update.","url":"https://blog.aifutures.org/feed"},{"date":"2026-08-01","kind":"milestone","lab":"anthropic","labName":"Anthropic","title":"Claude leads 26% of Anthropic's AI R&D work","text":"Up from under 1% in February 2026, by Anthropic's own measure; not yet fully autonomous.","url":"https://www.anthropic.com/institute/measuring-pace-of-ai-development"},{"date":"2026-07-24","kind":"record","lab":"anthropic","labName":"Anthropic","title":"ARC-AGI-3 record: 30.2%","text":"Claude Opus 5 (High effort)","url":"https://arcprize.org/results/anthropic-claude-opus-5"},{"date":"2026-07-24","kind":"record","lab":"anthropic","labName":"Anthropic","title":"Epoch Capabilities Index record: 162.94","text":"Claude Opus 5","url":"https://epoch.ai/data/eci_scores.csv"},{"date":"2026-07-09","kind":"record","lab":"openai","labName":"OpenAI","title":"ARC-AGI-2 record: 92.5%","text":"GPT-5.6 Sol (max effort)","url":"https://arcprize.org/results/openai-gpt-5-6-sol"},{"date":"2026-07-09","kind":"record","lab":"openai","labName":"OpenAI","title":"ARC-AGI-3 record: 7.78%","text":"GPT-5.6 Sol (max effort)","url":"https://arcprize.org/results/openai-gpt-5-6-sol"},{"date":"2026-05-18","kind":"forecast","lab":null,"labName":null,"title":"Polymarket: 14%","text":"Thin liquidity (about $9.1K visible). A market price, not evidence.","url":"https://www.orrery.me/markets/openai-announces-it-has-achieved-agi-before-2027"},{"date":"2026-04-23","kind":"record","lab":"openai","labName":"OpenAI","title":"Epoch Capabilities Index record: 162.38","text":"GPT-5.5 Pro","url":"https://epoch.ai/data/eci_scores.csv"},{"date":"2026-04-02","kind":"forecast","lab":null,"labName":null,"title":"AI Futures Project: Daniel Kokotajlo: Jun 2028","text":"Q1 2026 update, earlier from late 2029, citing METR Time Horizon 1.1 and newer models.","url":"https://blog.aifutures.org/p/q1-2026-timelines-update"},{"date":"2026-04-02","kind":"forecast","lab":null,"labName":null,"title":"AI Futures Project: Eli Lifland: Jun 2030","text":"Q1 2026 update, earlier from early 2032.","url":"https://blog.aifutures.org/p/q1-2026-timelines-update"},{"date":"2026-03-20","kind":"record","lab":"anthropic","labName":"Anthropic","title":"METR time horizon record: 12 hours","text":"Claude Opus 4.6 (current logistic model)","url":"https://metr.org/notes/2026-03-20-impact-of-modelling-assumptions-on-time-horizon-results/"},{"date":"2026-03-05","kind":"record","lab":"openai","labName":"OpenAI","title":"Epoch Capabilities Index record: 159.12","text":"GPT-5.4 Pro","url":"https://epoch.ai/data/eci_scores.csv"},{"date":"2026-02-05","kind":"record","lab":"openai","labName":"OpenAI","title":"Epoch Capabilities Index record: 156.82","text":"GPT-5.3 Codex","url":"https://epoch.ai/data/eci_scores.csv"},{"date":"2026-01-29","kind":"record","lab":"anthropic","labName":"Anthropic","title":"METR time horizon record: 5.333 hours","text":"Claude Opus 4.5 (Time Horizon 1.1 methodology)","url":"https://metr.org/blog/2026-1-29-time-horizon-1-1/"},{"date":"2025-12-31","kind":"forecast","lab":null,"labName":null,"title":"AI Futures Project: Daniel Kokotajlo: Oct 2029","text":"AI Futures Model, Dec 2025 update. The value is as restated in the Q1 2026 update (\"late 2029\"); the month is Laurenceo's reading of that phrase.","url":"https://blog.aifutures.org/p/ai-futures-model-dec-2025-update"},{"date":"2025-12-31","kind":"forecast","lab":null,"labName":null,"title":"AI Futures Project: Eli Lifland: Feb 2032","text":"AI Futures Model, Dec 2025 update. The value is as restated in the Q1 2026 update (\"early 2032\"); the month is Laurenceo's reading of that phrase.","url":"https://blog.aifutures.org/p/ai-futures-model-dec-2025-update"},{"date":"2025-12-11","kind":"record","lab":"openai","labName":"OpenAI","title":"Epoch Capabilities Index record: 155.41","text":"GPT-5.2 Pro","url":"https://epoch.ai/data/eci_scores.csv"},{"date":"2025-12-05","kind":"record","lab":"anthropic","labName":"Anthropic","title":"ARC-AGI-2 record: 37.6%","text":"Claude Opus 4.5 (Thinking, 64k)","url":"https://arcprize.org/blog/arc-prize-2025-results-analysis"},{"date":"2025-11-19","kind":"record","lab":"openai","labName":"OpenAI","title":"METR time horizon record: 2.667 hours","text":"GPT-5.1-Codex-Max","url":"https://metr.org/evaluations/gpt-5-1-codex-max-report/"},{"date":"2025-08-07","kind":"record","lab":"openai","labName":"OpenAI","title":"METR time horizon record: 2.283 hours","text":"GPT-5","url":"https://metr.org/evaluations/gpt-5-report/"},{"date":"2025-04-16","kind":"record","lab":"openai","labName":"OpenAI","title":"METR time horizon record: 1.5 hours","text":"OpenAI o3","url":"https://metr.org/evaluations/openai-o3-report/"},{"date":"2025-03-19","kind":"record","lab":"anthropic","labName":"Anthropic","title":"METR time horizon record: 1 hours","text":"Claude 3.7 Sonnet","url":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/"},{"date":"2024-01-05","kind":"forecast","lab":null,"labName":null,"title":"AI Impacts expert survey (2023): Jan 2047","text":"Survey of 2,778 published AI researchers, run in late 2023.","url":"https://arxiv.org/abs/2401.02843"},{"date":"2023-03-14","kind":"milestone","lab":"openai","labName":"OpenAI","title":"GPT-4 released","text":"A large step in general capability; the reference point for most early AGI forecasts.","url":"https://openai.com/index/gpt-4-research/"},{"date":"2022-11-30","kind":"milestone","lab":"openai","labName":"OpenAI","title":"ChatGPT launches","text":"Conversational AI reaches a mass audience.","url":"https://openai.com/index/chatgpt/"}],"changelog":[{"date":"2026-10-11","items":["Observatory opened: estimate Jun 2028, likely range Nov 2027 to Apr 2029, from 4 counted sources.","Counted: the three AI Futures Project forecasters and the AI task-length trend. Shown but not counted: a 2023 researcher survey (too old), OpenAI's own target and a Polymarket market.","Leaders are read from four frontier tests: METR task length, ARC-AGI-2, ARC-AGI-3 and the Epoch Capabilities Index."],"material":true},{"date":"2026-08-16","items":["Estimate held at Jun 2028; likely range now Nov 2027 to Apr 2029.","AI Futures Project: Daniel Kokotajlo now Nov 2027 (was Jun 2028).","AI Futures Project: Eli Lifland now Jan 2030 (was Jun 2030).","AI Futures Project: Brendan Halstead added: Jan 2029."],"material":true,"replayed":true},{"date":"2026-06-26","items":["Estimate held at Jun 2028; likely range now Oct 2027 to Jun 2030.","AI task-length trend (METR data) now Oct 2027 (was Sep 2027)."],"material":true,"replayed":true},{"date":"2026-04-02","items":["Estimate earlier by 16 months: Oct 2029 to Jun 2028.","AI Futures Project: Daniel Kokotajlo now Jun 2028 (was Oct 2029).","AI Futures Project: Eli Lifland now Jun 2030 (was Feb 2032)."],"material":true,"replayed":true},{"date":"2026-03-20","items":["Estimate held at Oct 2029; likely range now Sep 2027 to Feb 2032.","AI task-length trend (METR data) now Sep 2027 (was Apr 2028)."],"material":true,"replayed":true},{"date":"2026-01-29","items":["Estimate earlier by 14 months: Dec 2030 to Oct 2029.","AI Impacts expert survey (2023) no longer counted: older than 24 months.","AI task-length trend (METR data) now Apr 2028 (was Dec 2028)."],"material":true,"replayed":true},{"date":"2025-12-31","items":["Starting point: estimate Dec 2030, likely range Jul 2029 to Nov 2035, from 4 counted sources."],"material":true,"replayed":true}],"method":[{"term":"The estimate","text":"The middle (median) of the latest arrival dates from the counted sources: published forecasts whose definition is close to ours, plus one measured trend. Forecasts older than 24 months are shown but not counted."},{"term":"Likely range","text":"The middle half of the counted forecasts when there are four or more, otherwise all of them. It shows how far credible forecasters disagree, which is most of the uncertainty."},{"term":"AI task-length trend","text":"METR measures the length of software task (in human working time) that the best AI finishes half the time. We extend its record at its fitted doubling time and date the month it reaches one working month (167 hours) with at least even odds."},{"term":"Confidence","text":"Low with fewer than three counted sources or a likely range of two years or more, otherwise Medium. Never High: no AGI forecast can be checked until it happens."},{"term":"Leading company and country","text":"Most current records across four frontier tests (METR task length, ARC-AGI-2, ARC-AGI-3, Epoch Capabilities Index), then most records set in the last 12 months, then the most recent record. Models are attributed to labs by name."},{"term":"Not counted","text":"Company targets, prediction markets on announcements, old surveys and sources the daily pipeline adds are shown for context. They are counted only after a reviewed change, so one noisy source cannot move the estimate."},{"term":"History","text":"Replayed from the dated forecasts and measurements, using only what was published on each date, starting from the first day three sources could be counted. Since the observatory opened, each day's result is also stored and cannot be rewritten."},{"term":"Daily update","text":"Each morning the research step looks for new forecasts, frontier releases and benchmark records. Only items backed by a verified source are added; the estimate and leaders are then recomputed. Most days nothing changes, and the log says so."}],"summary":{"asOf":"2026-10-11","central":"2028-06","low":"2027-11","high":"2029-04","n":4,"confidence":"medium","centralYear":2028.46,"lowYear":2027.85,"highYear":2029.29,"centralLabel":"Jun 2028","lowLabel":"Nov 2027","highLabel":"Apr 2029","line":"Combining the leading published forecasts, AI that can do the work of AI researchers is most likely to arrive around Jun 2028, with the middle of the forecasts between Nov 2027 and Apr 2029.","company":{"id":"openai","name":"OpenAI","margin":"narrow","current":2,"runnerUp":"Anthropic"},"country":{"id":"US","name":"United States","margin":"clear"},"today":{"material":true,"text":"Observatory opened: estimate Jun 2028, likely range Nov 2027 to Apr 2029, from 4 counted sources."},"away":"about 1 year 8 months from now","moved":{"months":-30,"text":"2 years 6 months earlier than on 2025-12-31","span":"2 years 6 months","from":"2030-12","fromLabel":"Dec 2030","fromDate":"2025-12-31"},"href":"/intelligence/forecasts#agi"},"degraded":[]}