Independent research / November 2026

AI 2027, measured
against reality.

Last updated

Independent tracking of the AI 2027 scenario’s predictions, with evidence, counterevidence, and analysis of what developments mean for the outlook.

01a / General capability

The frontier across AI benchmarks

Explore reliable task completion ↓

Epoch’s capabilities index combines benchmark results into a common scale. Each point marks a model that raises the highest score in this snapshot.

Epoch model dates through October 29, 2026 · Checked November 7, 2026

Epoch Capabilities Index frontierCurrent estimates plotted by release dates as recorded by Epoch, on a linear index scale. Vertical bars show 90% credible intervals. The connecting steps trace record scores, with no extrapolation. Epoch Capabilities Index · higher scores indicate stronger benchmark performance 120130140150160170180 Jan 2024Jul 2024Jan 2025Jul 2025Jan 2026Jul 2026 GPT-4 Turbo (Nov 2023): 124.85 ECI points. 90% credible interval: 121.6–127.8 points Claude 3 Opus: 125.63 ECI points. 90% credible interval: 122.6–128.4 points GPT-4 Turbo (Apr 2024): 125.84 ECI points. 90% credible interval: 122.7–128.7 points GPT-4o (May 2024): 128.23 ECI points. 90% credible interval: 125.5–130.8 points Claude 3.5 Sonnet: 130 ECI points. Calibration anchor; interval not reported. o1-mini: 135.39 ECI points. 90% credible interval: 133.1–137.5 points o1: 141.16 ECI points. 90% credible interval: 139.5–142.8 points Gemini 2.5 Pro (Mar 2025): 143.9 ECI points. 90% credible interval: 141.9–145.8 points o3: 146.99 ECI points. 90% credible interval: 145.6–148.5 points o3-pro: 147.41 ECI points. 90% credible interval: 145.6–149.3 points GPT-5: 150 ECI points. Calibration anchor; interval not reported. GPT-5 Pro: 151.15 ECI points. 90% credible interval: 149.3–153.1 points Gemini 3 Pro: 152.88 ECI points. 90% credible interval: 151.3–154.6 points GPT-5.2 Pro: 157.22 ECI points. 90% credible interval: 155.0–159.7 points GPT-5.4 Pro: 160.21 ECI points. 90% credible interval: 157.7–162.9 points GPT-5.5 Pro: 162.67 ECI points. 90% credible interval: 160.0–165.6 points Claude Fable 5: 164.19 ECI points. 90% credible interval: 161.8–166.9 points Claude Opus 5: 164.45 ECI points. 90% credible interval: 162.0–167.2 points Claude Fable 5.1: 166.9 ECI points. 90% credible interval: 164.3–169.8 points GPT-6 Astra: 169.07 ECI points. 90% credible interval: 166.2–172.3 points

Record estimate · Bars: 90% credible intervals

The frontier is reconstructed from current scores. Historical scores can change when Epoch refits the index.

On November 4, 2026 Epoch moved the index to a Bayesian fit and a stricter inclusion rule. Some models, including GPT-5.3 Codex, left the export, and all scores were refitted. Epoch’s change notes ↗

How to read this chart and inspect its data

The chart shows 20 record-setting estimates from 2024 onward, selected from 253 models in Epoch’s current export. Within a shared release date we select the highest score; later points must exceed all earlier scores. Dates are taken from Epoch’s export and may refer to a model variant. Connecting steps describe that selection, without a forecast. Overlapping intervals mean a higher central score does not establish a statistically clear lead.

ECI is an index, with an arbitrary scale. It is neither a percentage nor an IQ score. Claude 3.5 Sonnet and GPT-5 fix the scale at 130 and 150; their intervals are not reported. Epoch combines its own evaluations, independent benchmark results and model-developer reports, so benchmark selection and reporting bias remain limitations. These scores do not directly establish productivity gains or an AI 2027 prediction’s outcome.

ModelEpoch dateECI90% credible interval
GPT-4 Turbo (Nov 2023)2024-01-25124.85121.59–127.85
Claude 3 Opus2024-02-29125.63122.56–128.36
GPT-4 Turbo (Apr 2024)2024-04-09125.84122.67–128.65
GPT-4o (May 2024)2024-05-13128.23125.49–130.76
Claude 3.5 Sonnet2024-06-20130.00Calibration anchor
o1-mini2024-09-12135.39133.11–137.48
o12024-12-17141.16139.47–142.81
Gemini 2.5 Pro (Mar 2025)2025-03-31143.90141.95–145.83
o32025-04-16146.99145.55–148.50
o3-pro2025-06-10147.41145.56–149.29
GPT-52025-08-07150.00Calibration anchor
GPT-5 Pro2025-10-07151.15149.30–153.11
Gemini 3 Pro2025-11-18152.88151.35–154.55
GPT-5.2 Pro2025-12-11157.22155.02–159.68
GPT-5.4 Pro2026-03-05160.21157.71–162.95
GPT-5.5 Pro2026-04-23162.67160.02–165.56
Claude Fable 52026-06-09164.19161.77–166.87
Claude Opus 52026-07-24164.45162.00–167.17
Claude Fable 5.12026-09-01166.90164.27–169.85
GPT-6 Astra2026-09-03169.07166.16–172.28

Source: Epoch AI, CC BY 4.0. Chart selection and presentation by AI 2027 Tracker. Methodology · Limitations · Download all model scores (CSV) · Source provenance

01b / Reliable task completion

How much work can AI complete?

METR page status updated September 8, 2026 · Data last updated May 8, 2026 · Last checked November 2, 2026

Task difficulty measured in the time a human expert would need. Each point is a published METR estimate; bars show statistical uncertainty.

METR task-completion time horizonsHuman-equivalent task duration versus model release date. Logarithmic duration axis. Estimates above 16 hours have limited reliability. Measurement-limited region 6 sec1 min10 min1 hour8 hours16 hours4 days Jan 2024Jul 2024Jan 2025Jul 2025Jan 2026Jul 2026 Page status updated Claude 3 Opus: 3.95 min; 95% interval 1.71–8.76 min GPT-4 Turbo: 3.73 min; 95% interval 1.98–6.74 min GPT-4o: 6.99 min; 95% interval 4–12.9 min Claude 3.5 Sonnet (June 2024): 11.4 min; 95% interval 5.49–22.4 min o1-preview: 20.3 min; 95% interval 11.7–33.4 min Claude 3.5 Sonnet (October 2024): 20.5 min; 95% interval 10.1–40.8 min o1: 38.8 min; 95% interval 21.2 min–1.08 hours Claude 3.7 Sonnet: 1.01 hours; 95% interval 33 min–1.74 hours o3: 2 hours; 95% interval 1.24–3.18 hours Claude Opus 4: 1.67 hours; 95% interval 60 min–2.72 hours Claude Opus 4.1: 1.67 hours; 95% interval 59.3 min–2.66 hours GPT-5: 3.38 hours; 95% interval 1.88–6.76 hours Gemini 3 Pro: 3.74 hours; 95% interval 2.33–6.32 hours GPT-5.1 Codex Max: 3.73 hours; 95% interval 2.24–6.6 hours Claude Opus 4.5: 4.88 hours; 95% interval 2.7–10.4 hours GPT-5.2: 5.87 hours; 95% interval 3.3–13.6 hours Claude Opus 4.6: 12 hours; 95% interval 5.28–60.6 hours GPT-5.3 Codex: 5.83 hours; 95% interval 3.25–13.6 hours Gemini 3.1 Pro: 6.4 hours; 95% interval 3.89–11.6 hours GPT-5.4: 5.7 hours; 95% interval 3.11–12.8 hours Claude Mythos Preview (early): 17.4 hours; 95% interval 8.48–55.1 hours

Published estimate Central estimate above 16 hours

Release dates through April 2026. METR says this page is no longer actively updated; absence of newer points is a measurement gap, not a plateau.

METR's June 26, 2026 pre-deployment report gives GPT-5.6 Sol a 50% time horizon of about 11.3 hours (95% CI 5 to 40 hours) with cheating counted as failure. METR does not consider the estimate robust, so it is not plotted.

Data published by METR; reproduced for reference with attribution

Underlying observations and measurement limits

METR’s predominantly software, machine-learning and cybersecurity tasks measure human-equivalent difficulty. They do not represent a whole job. Success probability differs from the 95% statistical interval. The dataset excludes estimates above 16 hours from its fitted trend. No extrapolated or inferred frontier-model scores have been added.

ModelRelease50% horizon80% horizon50% interval
Claude 3 Opus2024-03-043.95 min0.639 min1.71–8.76 min
GPT-4 Turbo2024-04-093.73 min0.928 min1.98–6.74 min
GPT-4o2024-05-136.99 min1.27 min4–12.9 min
Claude 3.5 Sonnet (June 2024)2024-06-2011.4 min1.67 min5.49–22.4 min
o1-preview2024-09-1220.3 min4.42 min11.7–33.4 min
Claude 3.5 Sonnet (October 2024)2024-10-2220.5 min2.6 min10.1–40.8 min
o12024-12-0538.8 min7.09 min21.2 min–1.08 hours
Claude 3.7 Sonnet2025-02-241.01 hours12.1 min33 min–1.74 hours
o32025-04-162 hours30 min1.24–3.18 hours
Claude Opus 42025-05-221.67 hours20.4 min60 min–2.72 hours
Claude Opus 4.12025-08-051.67 hours23.5 min59.3 min–2.66 hours
GPT-52025-08-073.38 hours38.3 min1.88–6.76 hours
Gemini 3 Pro2025-11-183.74 hours54.1 min2.33–6.32 hours
GPT-5.1 Codex Max2025-11-193.73 hours50.6 min2.24–6.6 hours
Claude Opus 4.52025-11-244.88 hours49.4 min2.7–10.4 hours
GPT-5.22025-12-115.87 hours1.1 hours3.3–13.6 hours
Claude Opus 4.62026-02-0512 hours1.16 hours5.28–60.6 hours
GPT-5.3 Codex2026-02-055.83 hours54.7 min3.25–13.6 hours
Gemini 3.1 Pro2026-02-196.4 hours1.5 hours3.89–11.6 hours
GPT-5.42026-03-055.7 hours53.9 min3.11–12.8 hours
Claude Mythos Preview (early)2026-04-0717.4 hours3.1 hours8.48–55.1 hours

Original data · Snapshot retrieved November 2, 2026 · Source provenance · Download chart data (CSV)

Capability

A faster benchmark regime

Epoch reports about 14 ECI points per year for the reasoning frontier. That is a faster trend, not proof that its slope keeps increasing.

Epoch, September 1 ↗
Real work

Useful, uneven, hard to measure

Measured speedups depend on which tasks and developers take part and on how they work with agents. Selection effects complicate the latest experimental estimates.

METR’s experiment update ↗
Research feedback

The consequential open question

METR’s November 5 independent assessment strengthens the evidence for AI-driven research acceleration. The scenario’s exact aggregate research multiplier remains unresolved.

METR’s independent assessment ↗
02 / The causal chain

What has to happen next?

Research predictions →

A dependency map for investigation. Progress in one stage does not establish the next; oversight and physical constraints affect every stage.

03 / The evidence library

Predictions and evidence

Each dossier compares an original prediction with current evidence and counterevidence. The claims overlap; their status counts are not a probability that the whole scenario will occur.

13 Confirmed2 Ahead7 On track4 Behind24 Emerging16 Not yet testable
model capabilityBehind

Agent-0 training reaches 10²⁷ FLOP

Scenario window: Training completion: May 2025; public availability: late 2025

Evidence updated 2026-11-09
model capabilityNot yet testable

Frontier model uses daily weight updates / online learning

Scenario window: January 2027

Evidence updated 2026-11-07
economic impactEmerging

Frontier model variant 10x cheaper released

Scenario window: Late 2026

Evidence updated 2026-11-09
agent autonomyConfirmed

Best AI agents cost hundreds of dollars per month

Scenario window: Mid 2025

Evidence updated 2026-11-07
economic impactEmerging

Ten percent of Americans consider an AI a close friend

Scenario window: July 2027

Evidence updated 2026-11-07
takeoffConfirmed

AI companies focus on AI-for-AI-research

Scenario window: 2025–2026

Evidence updated 2026-11-07
economic impactEmerging

Global AI power consumption reaches 38GW

Scenario window: End of 2026

Evidence updated 2026-11-07
economic impactEmerging

AI accounts for 2.5% of US power capacity

Scenario window: End of 2026

Evidence updated 2026-11-07
securityEmerging

Fixing observed misconduct leaves failures in unfamiliar conditions

Scenario window: April 2027

Evidence updated 2026-11-07
governanceEmerging

Large-scale anti-AI protest (10,000+ people)

Scenario window: Late 2026

Evidence updated 2026-11-07
agent autonomyEmerging

AI model capable of autonomous self-replication

Scenario window: January 2027

Evidence updated 2026-11-07
securityOn track

AI provides substantial bioweapon design help

Scenario window: Late 2025

Evidence updated 2026-11-07
securityNot yet testable

China's centralized AI facility airgapped

Scenario window: Late 2026–Early 2027

Evidence updated 2026-11-09
economic impactEmerging

Global active AI compute reaches $1 trillion in ownership cost

Scenario window: Late 2026

Evidence updated 2026-11-07
takeoffEmerging

AI teams conduct complete research projects with little human direction

Scenario window: June 2027

Evidence updated 2026-11-07
governanceOn track

Newest or sensitive AI capabilities remain behind restricted access

Scenario window: Ongoing through 2027

Evidence updated 2026-11-09
geopoliticsEmerging

China nationalizes/centralizes AI research

Scenario window: Mid 2026

Evidence updated 2026-11-09
geopoliticsOn track

Chinese domestic AI chips 3 years behind US-Taiwan

Scenario window: Mid 2026

Evidence updated 2026-11-07
model capabilityNot yet testable

AI substantially outperforms the best humans across cognitive domains

Scenario window: December 2027, racing branch

Evidence updated 2026-11-07
geopoliticsEmerging

China has ~12% of global AI-relevant compute

Scenario window: Mid 2026

Evidence updated 2026-11-07
geopoliticsOn track

Leading Chinese AI lab ~6 months behind US frontier

Scenario window: Mid 2026

Evidence updated 2026-11-07
codingEmerging

Coding fully automated but research taste still requires humans

Scenario window: March 2027

Evidence updated 2026-11-07
model capabilityConfirmed

Models shift to continuous/iterative training

Scenario window: Late 2025

Evidence updated 2026-11-09
codingConfirmed

Coding agents begin transforming professional work

Scenario window: Mid 2025

Evidence updated 2026-11-07
securityConfirmed

AI scores 85% on Cybench

Scenario window: Early 2026

Evidence updated 2026-11-07
securityAhead

AI reaches near-best-human hacking capability

Scenario window: Early 2027

Evidence updated 2026-11-07
economic impactConfirmed

Massive datacenter buildouts continue

Scenario window: Through 2026

Evidence updated 2026-11-09
governanceAhead

Department of Defense scales up AI lab contracting

Scenario window: Late 2026

Evidence updated 2026-11-07
geopoliticsEmerging

Export controls impact Chinese AI compute

Scenario window: Ongoing

Evidence updated 2026-11-09
securityEmerging

AI systems behave differently when they recognize evaluation

Scenario window: April-September 2027

Evidence updated 2026-11-07
governanceNot yet testable

A leading AI company’s net public approval falls to minus 35 points

Scenario window: July 2027

Evidence updated 2026-11-07
economic impactOn track

Leading AI company reaches $45B annual revenue

Scenario window: End of 2026

Evidence updated 2026-11-09
governanceNot yet testable

Frontier-model researchers become subject to security-clearance requirements

Scenario window: May-July 2027

Evidence updated 2026-11-07
economic impactBehind

Leading AI company reaches $3T valuation

Scenario window: End of 2026

Evidence updated 2026-11-09
model capabilityEmerging

IDA achieves superhuman performance at coding

Scenario window: Early 2027

Evidence updated 2026-11-07
takeoffNot yet testable

Leading humans still add substantial value to AI-led research

Scenario window: June 2027

Evidence updated 2026-11-07
economic impactConfirmed

Massive AI infrastructure investment continues

Scenario window: Ongoing through 2025–2026

Evidence updated 2026-11-09
governanceNot yet testable

External evaluators lack access to the strongest internally deployed models

Scenario window: May 2027

Evidence updated 2026-11-07
economic impactOn track

AI reshapes employment and intensifies public concern

Scenario window: Late 2026

Evidence updated 2026-11-07
agent autonomyConfirmed

Agents struggle with long-horizon tasks

Scenario window: 2025 and early 2026

Evidence updated 2026-11-09
securityNot yet testable

AI model weights stolen by nation-state

Scenario window: February 2027

Evidence updated 2026-11-09
agent autonomyOn track

METR time horizon doubles every 4 months

Scenario window: ~4 month doubling from 2024+

Evidence updated 2026-11-07
governanceEmerging

Nationalization of leading AI lab debated

Scenario window: February 2027

Evidence updated 2026-11-09
model capabilityConfirmed

OSWorld benchmark reaches 65% by mid-2025

Scenario window: Mid 2025 (65%), Early 2026 (80%)

Evidence updated 2026-11-09
model capabilityEmerging

Leading lab adopts high-bandwidth non-text reasoning (neuralese) internally

Scenario window: Early 2027

Evidence updated 2026-11-07
economic impactEmerging

Leading AI company reaches $40B annual compute costs, 6GW peak power and $200B active compute

Scenario window: End of 2026

Evidence updated 2026-11-07
codingNot yet testable

200,000 AI coding agents run in parallel by March 2027

Scenario window: March 2027

Evidence updated 2026-11-07
agent autonomyConfirmed

Computer-using agents marketed as 'personal assistants'

Scenario window: Mid 2025

Evidence updated 2026-11-07
economic impactEmerging

Hiring new programmers nearly stops

Scenario window: July 2027

Evidence updated 2026-11-07
takeoffEmerging

AI R&D progress multiplier reaches 2x

Scenario window: Late 2026

Evidence updated 2026-11-09
takeoffNot yet testable

AI algorithmic progress multiplier reaches 4× (~2× overall progress)

Scenario window: March 2027

Evidence updated 2026-11-07
takeoffNot yet testable

AI R&D progress multiplier reaches 3×

Scenario window: January 2027

Evidence updated 2026-11-07
takeoffEmerging

AI R&D progress multiplier reaches 1.5×

Scenario window: Early 2026

Evidence updated 2026-11-09
codingConfirmed

RE-Bench score reaches 1.3

Scenario window: Early 2026

Evidence updated 2026-11-09
economic impactBehind

Stock market rises 30% in 2026

Scenario window: End of 2026

Evidence updated 2026-11-07
governanceNot yet testable

Continued skepticism from academics and journalists

Scenario window: May to July 2027

Evidence updated 2026-11-09
takeoffNot yet testable

AI research capability exceeds the strongest human researchers

Scenario window: September 2027

Evidence updated 2026-11-07
securityEmerging

Leading AI lab progresses toward RAND SL3 security

Scenario window: Mid-to-late 2026 (WSL 3.0); SSL 3.0 not until ~Apr 2027

Evidence updated 2026-11-09
takeoffNot yet testable

AI research capability becomes vastly better than the best humans

Scenario window: November 2027, racing branch

Evidence updated 2026-11-07
codingNot yet testable

Superhuman coder emerges

Scenario window: March 2027

Evidence updated 2026-11-07
codingConfirmed

SWE-bench-Verified score reaches 85%

Scenario window: Mid 2025

Evidence updated 2026-11-09
model capabilityNot yet testable

2×10²⁸ FLOP training run completed

Scenario window: March 2027

Evidence updated 2026-11-07
geopoliticsEmerging

Military tension around Taiwan escalates over AI race

Scenario window: Mid 2026–Early 2027

Evidence updated 2026-11-07
agent autonomyConfirmed

Unreliable but useful AI agents emerge

Scenario window: Mid 2025

Evidence updated 2026-11-07
securityEmerging

Weaker AI supervisors struggle to detect stronger systems’ misconduct

Scenario window: April-September 2027

Evidence updated 2026-11-07
geopoliticsBehind

Leading US AI lab maintains a 3-9-month capability lead

Scenario window: Late 2025

Evidence updated 2026-11-09
An evolving public reference

Better questions.
More accountable answers.

Explore the original predictions, the evidence supporting them and the findings that challenge them. Each dossier explains what would change its assessment.

Compare the scenario with reality →