AGI Scorecard中文
All sections

Make sense of AI. Find your next step.

Start with a perspective worth watching. Explore what AI means for work, learning and the future.

Explore videos & ideas ↗
720 videos312 audio episodes30 reviewed views

Follow the questions that matter to you.

Interviews, creator videos and research. Choose a perspective.

Browse all videos ↗

Daily English picks

See more ↗

A fresh selection each day · · 00:00 UTC+8

Checked · Daily source checks; reviewed summaries are published separately.

Turn a good question into a next step.

Save a perspective, inspect the evidence, or give Jarvis a question to investigate.

Check the future against the evidence.

Track 8 predictions from Situational Awareness: what has held up, and what remains open.

Inspect the evidenceMake my assessment

Latest evidence notes

— Clarified the open-weight model gap: January–May 2026 study window, about 4 months on average; about 6 under a stricter comparison.

Latest verdict ledger remains dated 2026-09-06. This is a source-note and presentation update, not a new verdict.

It summarizes 8 Situational Awareness predictions in our published ledger. The 2026-09-06 reading is 62.5/100: 3 supportive, 4 unresolved and 1 refuted. It is not a general benchmark of all AI capabilities.

Cite this ledger

AGI Scorecard, “AGI-2027 Thesis Tracker,” ledger dated 2026-09-06, 62.5/100. https://agiscorecard.com/data.json

Reuse with attribution under CC BY 4.0. Preserve the reading date and the uncertainty of each verdict.

The Scorecard — Every Prediction vs Reality

Source: Situational Awareness (June 2024) · Ledger reading dated 2026-09-06 · Tap any row for evidence & flip conditions
Target Prediction Reality check (2026) Verdict
2025/26 Models outpace college graduates across knowledge work Frontier models at ~83% on knowledge-work benchmarks; agent products in production ✓ On track▶

Evidence ~83% on GDPval-style knowledge work and ~80% on SWE-Bench Pro; agent products in production. Drop-in reliability without supervision still lags the benchmarks.

Flips if Capability plateaus below skilled-graduate level, or production agent adoption stalls on reliability.

Sources GDPval (OpenAI) · SWE-bench

Full verdict: Can AI replace knowledge workers? →

~0.5 OOM/yr Compute + algorithmic scaling continues at trend One-year retrospectives find the pace roughly supported by evidence ✓ Holding▶

Evidence Effective compute has roughly held the ~0.5 OOM/yr pace; gains increasingly come from reasoning, tools and agents (unhobbling), not just raw scale.

Flips if A sustained multi-year drop below ~0.5 OOM/yr — most plausibly via a capex pullback.

Sources Epoch AI — Trends in AI

Full verdict: Is AI compute still scaling? →

$500B/yr Massive AI capex acceleration Investment exceeding projections; accelerating faster than forecast ✓ Exceeded▶

Evidence Investment exceeded the essay's projections — his most vindicated call. The caveat: revenue lags the spend, which is the bear case to watch.

Watch The revenue-vs-capex gap — a funding pullback would slow the compute trend for financial reasons.

Sources Epoch AI — Finances of AI

Full verdict: Did AI capex hit trillion-dollar scale? →

— Open-source fades; proprietary algorithms create a durable US moat DeepSeek V4 / Qwen open-weight models within 3–6 months of frontier ✗ Wrong▶

Evidence DeepSeek and Qwen open-weight models sit within ~3–6 months of the frontier at a fraction of the cost — the opposite of fading.

Flips back if The frontier pulls a multi-year lead via advances that don't diffuse. No sign of that yet.

Sources Epoch AI

Full verdict: Did open-source AI fade? →

2027 AGI: models do the work of an AI researcher/engineer Agentic coding strong (~80% SWE-Bench Pro) but autonomous research unconfirmed ⟳ Open▶

Evidence Agentic coding is strong (~80% SWE-Bench Pro) but no system has autonomously conducted AI research end-to-end — the bar that defines this claim.

Resolves By January 1, 2028. Fulfilled if autonomous AI research is demonstrated; Wrong if the deadline passes without it.

Full verdict: Will AGI arrive by 2027? →

2027/28 US government launches formal AGI project National-security involvement growing; no formal Project announced ⟳ Open▶

Evidence National-security involvement is growing (export controls, lab security requirements, defense interest) but no centralized, Manhattan-Project-style effort exists.

Resolves By 2027/28. The open-source verdict weakens its premise: diffused capability undercuts the case for a single Project.

Full verdict: Will the US government build AGI? →

2027–29 Intelligence explosion: decade of progress in 1 year Too early to grade ◌ Pending▶

Why pending Its trigger — AI autonomously doing AI research — hasn't been demonstrated, and the 2027–29 window hasn't arrived. Not enough evidence to grade either way.

Full analysis: Will there be an intelligence explosion? →

2030s Superintelligence; decisive geopolitical advantage Too early to grade ◌ Pending▶

Why pending Downstream of AGI and an intelligence explosion, neither of which has occurred. The furthest-out claim on the scorecard.

Full analysis: Will there be superintelligence? →

Source: situational-awareness.ai · Verdicts based on public retrospectives & benchmarks

The 2027 clock is ticking.

Get each verdict change the week it happens — free, no hype, just signal.

Subscribe free →

How would you grade these 8 predictions?

Choose delivered, unresolved or failed for each prediction. See your score and where you differ from the ledger. No account needed.

Equal weights: delivered = 1, unresolved = 0.5, failed = 0. The score is their mean × 100.

What did AI actually change? Let’s discuss.

Share a tool test, challenge an AGI forecast or compare how work is changing. Read freely; join to reply and follow.

AI tools in practiceWhich AI workflow survived a real task?The future of AIWhat would change your AGI timeline?AI and your workWhich part of your job actually changed?

New community. These are editorial opening prompts; member submissions appear after review.

Explore discussionsJoin free / my discussions
Try: AGI 2027 AI stocks job risk Buffett holdings red team
2026-09-28 · Gross historical simulation
What does the historical AI-sleeve simulation show?

Next-session closing prices after disclosure, equity-only holdings and complete price coverage. Historical gross returns exclude costs; they are not live profits or proof of future returns.

+179.5%
Druckenmiller AI slice / matched QQQ +55.5% / Buffett +34.6%
2024-08-15 → 2026-08-17
Inspect the method and period records →
📊 Research
Thesis Tracker · 62.5/100 The full scorecard All AGI questions 💼 Should you worry about AI & your job? ⏱ Amodei's “bloodbath” clock 🎵 Is AI music good enough yet? 97% can't tell — and still don't listen 🔥 Amodei says slow down — does it move the 2027 odds? ⚖️ EU AI Act: applies vs. deferred 🔐 Code Word: can you spot an AI voice clone? ⚡ Is AI raising your electricity bill? Agent protocols: really used? 📅 Prediction receipts 🎯 Calibration report card 🧪 Revenue experiment log
🎮 Play
🎲 The Future Bet Ecommerce Reporting Coach: practice a report and check the result ↗ Relay: make a story your friends can change 🎯 What's your AGI type? 📋 Will AI take my job? 🏆 Forecaster leaderboard 🕶 Matrix odds calculator 📈 AI stock exposure check 🎮 Games Ledger: is it dead? 🐸 Gridlings: 11 daily logic puzzles 🧰 All free tools →
💰 Invest
AI Investing Hub AI infrastructure: annual facts & research notes 📡 SunWatch: live market triggers 🏆 Did copying them actually pay? 🧭 Compass: legendary investors' 13F moves 🧨 The fund collapse, graded 持仓收据 · The receipts (中文) Is AI capex a bubble? Sponsor this site
🇨🇳 中文
中文主页 · 预言 vs 持仓 🎲 押注未来 💰 AI 投资板块 🎯 AGI 类型测试 🧰 免费工具合集
2019
GPT-2 — Preschooler level
Could barely string together coherent paragraphs. The baseline that showed language models had potential.
✓ Reached
2023
GPT-4 — Smart high schooler
Aces AP exams, writes sophisticated code, reasons through competition math. Benchmarks rapidly saturated.
✓ Reached
2024
PhD-level benchmarks cracked
GPQA (expert PhD questions) being approached. Claude 3 Opus at ~60% vs 80% for in-domain PhDs.
✓ In progress
2025–26
Outpace college graduates
Models able to perform at the level of a skilled recent grad across most professional tasks. Agent capabilities mature.
⟳ Approaching
2027
AGI — AI researcher level
Models capable of doing the work of an AI researcher/engineer. Could automate AI research itself, triggering intelligence explosion.
⟳ Forecast
2027–29
The Project begins
USG/national security involvement. Government AGI project launches. No startup handles superintelligence alone.
◌ Future
2029–30
Intelligence explosion
Hundreds of millions of AGIs automating AI research. A decade of progress compressed into 1 year.
◌ Forecast
2030s
Superintelligence
Vastly superhuman AI systems. Decisive economic and military advantage. The free world's survival at stake.
◌ Forecast
2030s
Trillion-dollar clusters
US electricity production grows tens of percent. Hundreds of millions of GPUs. Industrial mobilization unlike anything since WWII.
◌ Forecast
OOM Tracker — Effective Compute Scaleup
Orders of magnitude vs GPT-2 baseline
"With each OOM of effective compute, models predictably, reliably get better. If we can count the OOMs, we can extrapolate capability improvements."
— Leopold Aschenbrenner, Situational Awareness (2024)
Raw compute
~2 OOMs/yr
Algorithmic efficiency
~0.5 OOMs/yr
Unhobbling gains
~1 OOM total
GPT-2 → GPT-4 total
~5 OOMs
GPT-4 → 2027 (proj.)
+5 OOMs

Source: Epoch AI public estimates + Aschenbrenner analysis. Dashed bar = projection.

Frontier Labs — Status Overview
Best available public data · Updated June 2026
OpenAI
🇺🇸 USA
Flagship modelGPT-5.5 / 5.5 Pro
Annualized revenue$25B+ (est.)
NotableLeads FrontierMath T4
Release cadence~6 weeks/flagship
Anthropic
🇺🇸 USA
Flagship modelClaude Fable 5 (Mythos)
Annualized revenue~$19B (est.)
NotableNew tier above Opus
StanceSafeguarded frontier
Google DeepMind
🇺🇸 USA
Flagship modelGemini 3.1 Pro / 3.5
NotableGemini Spark agent
AdvantageTPU + multimodal lead
Context window2M tokens native
Chinese Labs (est.)
🇨🇳 China
Key playersDeepSeek, Qwen, MiniMax
Flagship modelsDeepSeek V4, Qwen 3.7 Max
StrategyOpen-weight + low cost
Gap vs frontier~3–6 months (est.)

His 2027 vs everyone else

Public AGI timelines · Mid-2026
Forecaster AGI timeline Camp
Elon Musk (xAI) By end of 2026 Most aggressive
Leopold Aschenbrenner 2027 "strikingly plausible" This scorecard
Demis Hassabis (DeepMind) ~50% by 2030 Lab leader, cautious
Samotsvety (pro forecasters) ~28% by 2030 Best track record
Metaculus community (~2,000 forecasters) 25% by 2029 · 50% by 2033 Crowd consensus
Andrej Karpathy (ex-OpenAI) ~A decade out Architecture skeptic
AI researcher survey (2,778 respondents) 50% by 2047 Academic median

Definitions of "AGI" vary by forecaster — direct comparison is approximate. Notably, expert medians have compressed from ~2060 to ~2033 in six years.

Recent site changes

Changes to features and evidence presentation. Verdict dates remain attached to the ledger.

  1. Choose an action plan from dropdowns, with an instant preview
  2. Bilingual homepages now open directly into videos and perspectives
  3. Four evidence topics now have dated citation notes and downloadable charts
Full change history

Frequently asked questions

What does the AGI Thesis Tracker measure?

It summarizes 8 Situational Awareness predictions in our published ledger. The 2026-09-06 reading is 62.5/100: 3 supportive, 4 unresolved and 1 refuted. It is not a general benchmark of all AI capabilities.

Does 62.5/100 mean a 62.5% probability of AGI?

No. Each verdict has equal weight: supportive = 1, unresolved = 0.5 and refuted = 0. The mean is multiplied by 100. This is an editorial composite, not a probability forecast or a finding that AGI has arrived.

How can I reproduce or cite the score?

Read predictions and thesisTracker in data.json, recompute the published weights, and inspect index-history.json for past readings. Cite AGI Scorecard, the ledger date 2026-09-06 and the original data URL. The dataset is CC BY 4.0; a site redesign date is not an evidence date.

Does the AGI countdown reaching zero mean AGI has arrived?

No. It counts down to January 1, 2027 at 00:00 UTC, the start of the observation window, not a promised AGI arrival. Our deadline for judging the by-end-of-2027 prediction is January 1, 2028, and still requires public evidence.

How can I vote on when AGI arrives?

Choose a time window below the homepage countdown to see your prediction type and historical answers, then copy, share or save your prediction on this device. No account is needed. Counts are answers, may repeat, and are not unique people or a representative survey.

📋 All AGI questions, answered → Was Aschenbrenner right? Will AGI arrive by 2027? Did open source fade? Aschenbrenner vs Metaculus All predictions tracked When will AGI arrive? Who is Aschenbrenner? Situational Awareness summary Aschenbrenner vs Hassabis Will the US govt build AGI? Intelligence explosion 2027? Superintelligence in the 2030s? Is AI compute still scaling? Trillion-dollar AI capex? Can AI replace knowledge workers? Will AI cause mass unemployment? What jobs are safe from AI? Will AI take over? Is AGI inevitable? Aschenbrenner vs Musk Is China beating the US to AGI? Situational Awareness vs AI 2027 Is AGI just hype? Will AI replace programmers? Hassabis's AGI prediction Musk's AGI prediction Is the AI capex a bubble? What is superintelligence? AI orders of magnitude What is unhobbling? How close are we to AGI? DeepSeek vs OpenAI What is AI 2027? US–China AI arms race What is GDPval? Karpathy's AGI prediction 🏆 Who's winning the AGI bet? 📅 AI prediction receipts 📋 Will AI take my job? 60-sec check 💰 AI Investing Hub: who's betting what 🎲 The Future Bet: what do you bet happens? What is SWE-Bench? AI progress in 2026 so far Are scaling laws dead? What is AGI? Is ChatGPT AGI? AI vs human intelligence How fast is AI improving? Narrow vs general AI How will we know AGI arrived? What is the singularity? Altman's AGI prediction Amodei's AGI prediction AGI vs superintelligence The full timeline, graded Altman vs Musk on AGI Altman vs Amodei Musk vs Hassabis Karpathy vs Altman Data for AI agents

Explore the AI-agent stack

Plain-English references and free tools we build for the AI-agent era.

🔍 Justin Sun's 2013 Bitcoin story — how much is verified? 🤖 What is B.AI? Sun's AI relay, audited 📉 Do AI trading agents make money? 📈 Is Nvidia overvalued? What its AI bet rides on 🧾 Does copying 13F filings work? We tested it 🧠 Is Claude getting dumber? 🧾 Evidence Audits — our method, for hire 🎲 The Paradigm Bet — a public pre-registered experiment 📖 Agent Glossary — 25 AI & agent definitions 🛠 Free tools — schema, robots.txt, sitemap, MCP config ⚡ MCP Pulse — health-check an MCP server 🛒 AgentReady — is your store visible to AI shoppers?

Be first to know when the score moves

One email when a verdict flips or the AGI-2027 Thesis Tracker moves — the single auditable score no other tracker has. No hype, just signal.

Subscribe free →