Rolling 12 months · Claude Opus 5.5 + Perplexity
Will AGI be achieved in the next 12 months?
Updated daily. Each point is a new forecast with fresh research—not a forecast of the same fixed deadline. These forecasts are never graded and do not enter the benchmark.
Latest chance of YES
22%
Change from previous
+2.0 pp
Last forecast
Oct 22, 2026
Horizon ends Oct 22, 2027
Probability over time
Working definition
An AI demonstrates human-level performance across a broad range of cognitive tasks, supported by independent public evidence. A single benchmark result or an unsupported developer claim is not sufficient.
Latest forecast
Capabilities are advancing quickly: ARC-AGI-3 scores are near the ceiling, METR reports agents completing multi-day software projects, and Altman expects an internally declared AGI by year-end. However, the resolution needs independent public evidence of broad human-level cognition. No accepted body or procedure exists to certify that, and METR still finds weaknesses in strategic judgment. Hassabis puts AGI 2-3 years out, and contested definitions usually leave such questions unresolved within a short window. There is also a risk that AGI is judged to have been reached before the window opens.
Research dossier and forecast criteria
For this forecast, YES means AGI is first achieved after 2026-10-22 and on or before 2027-10-22. An AI demonstrates human-level performance across a broad range of cognitive tasks, supported by independent public evidence. A single benchmark result or an unsupported developer claim is not sufficient.
Research by Perplexity Sonar Searches run: October 2026 AGI human level AI broad cognitive tasks independent evidence latest | 2026 AI company AGI definition timeline statements OpenAI Anthropic Google DeepMind | 2026 AI benchmark evaluations limitations human-level capability independent assessment METR researchers | ARC Prize 2026 independent evaluation frontier AI model ARC-AGI results October 2026 | METR 2026 time horizon report AI agents human expert task duration October 2026 | Reuters 2026 AGI OpenAI Sam Altman AGI prediction 2026 2027 Anthropic Amodei | site:time.com Sam Altman AGI end of 2026 August 2026 internal system | Dario Amodei powerful AI late 2026 early 2027 official statement 2026 Anthropic | Demis Hassabis October 2026 AGI two to three years Wall Street Journal Magazine | METR June 2026 progress report time horizon agents software projects weeks primary report | Google DeepMind Levels of AGI paper definition performance breadth autonomy 2025 primary source | AI Impacts 2023 survey 2778 researchers 50 percent high level machine intelligence 2047 paper --- BRIEFING --- ## State of play There is no single agreed operational definition of AGI. Google DeepMind’s framework separates capability depth from breadth and autonomy, and says that measuring progress requires benchmarks capable of assessing model behavior across tasks.[39] Its public definition describes AGI as AI “at least as capable as humans at most cognitive tasks.”[38] The question’s additional requirement—broad human-level performance supported by independent public evidence—therefore sets a specific evidentiary bar; a lab’s own label or one benchmark score alone would not settle it. Recent independent benchmark results show substantial performance on some tests, but do not by themselves establish broad human-level cognition. ARC Prize’s results page, updated October 21, lists high scores for several frontier models on ARC-AGI-1 and ARC-AGI-2; for example, Gemini 4 Argon is listed at 98.5% and 91.3%, respectively.[11] On ARC-AGI-3, the same page lists Argon at 97.46% and GPT-6.1 Sol at 96.37%.[11] Those are results on particular benchmark suites, not evidence in themselves of human-level performance across a broad range of cognitive tasks. Other evidence measures different abilities. METR’s March–June 2026 report says agents can complete some software projects that would take human experts days or weeks, while describing weaker results on challenges involving strategic judgment, stealth, and modeling adversaries.[32] METR’s “time horizon” measure concerns the duration of tasks at which a model succeeds at a specified reliability; it is not a general-purpose AGI certification.[1] The sources reviewed do not identify a publicly accepted process or independent panel empowered to declare that AGI has been achieved. ## Actors, mechanics and timetable There is no formal regulatory vote or filing that determines whether AGI has been achieved under the question’s criteria. The practical question is whether public evidence emerges during the specified window, and whether it demonstrates broad human-level performance with independent support. The public positions of leading developers use differing thresholds. Anthropic’s submission to the U.S. Office of Science and Technology Policy says it expects “powerful AI systems” in late 2026 or early 2027, defining them in part by intellectual capabilities matching or exceeding Nobel Prize winners across most disciplines.[22] In a separate statement, Dario Amodei described a “country of geniuses in a datacenter” as possible by 2026 or 2027.[29] These are company-leader expectations about powerful AI, not an independent finding that the question’s AGI standard has been met. In an August 2026 interview, OpenAI CEO Sam Altman said the company was “not quite yet” at AGI, but expected that by year-end it would have an internal system he would call AGI; OpenAI’s charter definition is systems that outperform humans at most economically valuable work.[28] That is an internal expectation and uses a different definition from the resolution criteria. Google DeepMind’s Demis Hassabis was reported on October 21 as putting AGI two to three years away, while saying he was “updating nearer and nearer.”[27] The cited account does not describe a formal public test or review mechanism attached to that estimate. ## Signals and constraints The benchmark record offers both rapid progress and reasons to treat individual scores cautiously. ARC Prize’s results show strong scores across several versions of its tests, but the scores remain measurements on specific tasks.[11] METR reports notable performance on some long software tasks alongside limitations on strategic challenges, illustrating that capability varies by task type.[32] ARC Prize’s 2026 competition page gives December 4, 2026, as the date results are to be announced; those results may add evidence, but a competition outcome would still need to be evaluated against the broader resolution criteria.[14] Published timelines also vary by definition. A 2023 survey of 2,778 AI researchers reported an aggregate 50% date of 2047 for high-level machine intelligence, defined as unaided machines accomplishing every task better and more cheaply than human workers.[31][36] That survey predates the latest reported model results and uses a standard different from the company leaders’ descriptions; it is context, not a direct forecast of this question’s outcome. The sources reviewed do not provide an independently administered, comprehensive assessment establishing human-level performance across a broad range of cognitive tasks, nor do they specify a widely accepted procedure for conducting one. That evidentiary gap is central to the stated resolution criteria. --- SOURCES --- [1] How are countries evaluating AI models? Source: brookings.edu Published: 2026-10-16 Last updated: 2026-10-17 URL: https://www.brookings.edu/articles/how-are-countries-evaluating-ai-models/ Model Evaluation and Threat Research (METR) measures a model’s “time horizon,” the length of task, measured by how long human experts take to complete it, at which a model succeeds at a given rate of reliability. [11] Results - ARC Prize Source: arcprize.org Published: unknown Last updated: 2026-10-21 URL: https://arcprize.org/results # Results ### Model performance across ARC-AGI benchmarks DateProviderModelARC-AGI-1ARC-AGI-2ARC-AGI-3 Oct 17, 2026DeepSeekDeepSeek V4.1 Flash 101794.5%78.1%— Oct 16, 2026ReflectionBeam 1.074.0%23.5%— Oct 15, 2026Meta AIMuse Spark Open89.0%57.5%— Oct 15, 2026NVIDIANemotron 3.5 Super VL22.7%1.9%— Oct 13, 2026GoogleGemini 4 Argon98.5%91.3%97.46% Oct … [14] ARC Prize 2026 Source: arcprize.org Published: unknown Last updated: 2026-10-02 URL: https://arcprize.org/competitions/2026 #### Key Dates - **December 4, 2026** - Results announced ... ### Spirit of ARC Prize 2026 While anyone is welcome to use our public data and leaderboards independently, the privilege of competing officially - and benefiting from our private evaluations, global visibility, and chance to win prizes - comes with a commitment to transparency and openn… [22] Anthropic's Recommendations to OSTP for the U.S. AI ... Source: anthropic.com Published: 2025-03-06 Last updated: 2026-10-21 URL: https://www.anthropic.com/news/anthropic-s-recommendations-ostp-u-s-ai-action-plan # Anthropic’s recommendations to OSTP for the U.S. AI action plan In response to the White House’s Request for Information on an AI Action Plan, Anthropic has submitted recommendations to the Office of Science and Technology Policy (OSTP). ... As our CEO Dario Amodei writes in ‘Machines of Loving Grace’, we expect powerful AI systems will emerge in… [27] Google DeepMind's Demis Hassabis thinks AGI is two to three years away. Source: theverge.com Published: 2026-10-21 Last updated: 2026-10-21 URL: https://www.theverge.com/ai-artificial-intelligence/1015277/google-deepminds-demis-hassabis-thinks-agi-is-two-to-three-years-away Posted Oct 21, 2026 at 12:43 PM UTC Google DeepMind's Demis Hassabis thinks AGI is two to three years away. The Wall Street Journal published a big profile of the Google DeepMind co-founder, who says his timeline for the arrival of artificial general intelligence (AGI) is becoming 'nearer and nearer.' He adds that 'winning' the AI race 'means helpi… [28] ?_bhlid=b6cb46324d648ccd4f57c090a289ba926553fd42 Source: time.com Published: unknown Last updated: 2026-08-27 URL: https://time.com/article/2026/08/26/openai-sam-altman-interview/?_bhlid=b6cb46324d648ccd4f57c090a289ba926553fd42 ## Alex Heath Amid all this, company leaders believe they have arrived at the cusp of a milestone that could change the course of humanity: the creation of [artificial general intelligence](https://time.com/7312305/agi-race-us-china-trump/), or AGI. OpenAI’s charter defines AGI as “highly autonomous systems that outperform humans at most economical… [29] Statement from Dario Amodei on the Paris AI Action Summit Source: anthropic.com Published: 2025-02-11 Last updated: 2026-10-21 URL: https://www.anthropic.com/news/paris-ai-summit Dario Amodei on the Paris AI Action Summit \ Anthropic # Statement from Dario Amodei on the Paris AI Action Summit ... - Announcements Aug 4, 2026 ... We were pleased to attend the AI Action Summit in Paris, and we appreciate the French government’s efforts to bring together AI companies, researchers, and policymakers from across the world. ... Tim… [31] 2023 Expert Survey on Progress in AI - AI Impacts Wiki Source: wiki.aiimpacts.org Published: unknown Last updated: 2026-10-03 URL: https://wiki.aiimpacts.org/ai_timelines/predictions_of_human-level_ai_timelines/ai_timeline_surveys/2023_expert_survey_on_progress_in_ai # 2023 Expert Survey on Progress in AI The 2023 Expert Survey on Progress in AI is a survey of 2,778 AI researchers that AI Impacts ran in October 2023. ... ## Details ### Survey methods #### Participants We received 2,778 responses (where the person answered at least one question), for a response rate of 15%. ... ### Results #### Timing of human-l… [32] Progress Report March - June 2026 - metr.org Source: metr.org Published: unknown Last updated: 2026-09-02 URL: https://metr.org/june-2026-progress-report.pdf March - June 2026Model Evaluation and Threat Research Focus of this update Report Frontier Risk Report X thread Website METR published its first Frontier Risk Report, the most rigorous independent assessment of catastrophic risks from AI development to date. ... Pages 11-21 METR found that AI agents conducted significant engineering work at AI comp… [36] Survey: Median AI expert says 5% chance of human extinction ... Source: aiimpacts.org Published: unknown Last updated: 2026-08-14 URL: https://aiimpacts.org/wp-content/uploads/2024/01/EMBARGOED_-AI-Impacts-Survey-Release-Google-Docs.pdf from AI BERKELEY, CALIFORNIA: In a new survey of 2,778 AI experts, experts gave a median 5% chance that AI would cause human extinction. In the survey conducted by AI Impacts, a Berkeley-based think tank, in collaboration with researchers at the University of Bonn and University of Oxford, experts offered a median response of 5% for the … [38] Taking a responsible path to AGI Source: deepmind.google Published: 2025-04-02 Last updated: 2026-10-02 URL: https://deepmind.google/blog/taking-a-responsible-path-to-agi/ Taking a responsible path to AGI — Google DeepMind April 2, 2025 Responsibility & Safety ... # Taking a responsible path to AGI Artificial general intelligence (AGI), AI that’s at least as capable as humans at most cognitive tasks, could be here within the coming years. Integrated with agentic capabilities, AGI could supercharge AI to understand, r… [39] Levels of AGI for Operationalizing Progress on the Path to AGI Source: deepmind.google Published: 2024-07-21 Last updated: 2026-10-03 URL: https://deepmind.google/research/publications/66938/ Levels of AGI for Operationalizing Progress on the Path to AGI — Google DeepMind ... ## Unlocking a new era of discovery with AI July 21, 2024 # Levels of AGI for Operationalizing Progress on the Path to AGI ... We propose a framework for classifying the capabilities and behavior of Artificial General Intelligence (AGI) models and their precursors.…
Daily forecast history (2)
Oct 22, 2026 · 22% YESthrough Oct 22, 2027
Capabilities are advancing quickly: ARC-AGI-3 scores are near the ceiling, METR reports agents completing multi-day software projects, and Altman expects an internally declared AGI by year-end. However, the resolution needs independent public evidence of broad human-level cognition. No accepted body or procedure exists to certify that, and METR still finds weaknesses in strategic judgment. Hassabis puts AGI 2-3 years out, and contested definitions usually leave such questions unresolved within a short window. There is also a risk that AGI is judged to have been reached before the window opens.
Oct 21, 2026 · 20% YESthrough Oct 21, 2027
Frontier progress is very fast. Top models now post near-ceiling ARC-AGI-1/2/3 scores, and OpenAI leaders say they will have an internal system they call AGI by end of 2026; an OpenAI declaration verified by the Microsoft expert panel could plausibly count. Against that, capability remains jagged (clock reading, paper replication under 20%), no agreed test or certifying body exists, and the criterion requires independent public evidence of broad human-level performance rather than developer claims. Given the ambiguity and the high bar, a YES resolution within 12 months is possible but not the base case.