AGI Clock

Tracking humanity's last invention — in real time.

Open-source landing and dashboard that estimates AI progress toward Artificial General Intelligence. Shows AGI Readiness (0–100%) as a benchmark-based proxy across reasoning, engineering, autonomy, tool use, search, and reliability signals.

---

How it works

1. Daily collector (scripts/collect.mjs) fetches frontier scores for tracked benchmarks from the llm-stats.com API
2. Calculates a weighted AGI Readiness proxy and AGI Distance
3. Updates src/data/live.json and commits to the repo via GitHub Actions
4. GitHub Pages deploys the updated static site from the dist artifact

Benchmarks tracked

Dimension	Weight	Included signals
Reasoning + Expert Knowledge	30%	ARC-AGI v2, Humanity's Last Exam, GPQA, MMLU-Pro, FrontierMath
Coding + Engineering	25%	SWE-Bench Verified, SWE-Bench Pro, LiveCodeBench v6, Terminal-Bench, Terminal-Bench 2.1
Agents + Autonomy	30%	OSWorld, APEX-Agents, VITA-Bench, DeepPlanning, OfficeQA Pro, Finance Agent v2
Tool Use + Search Reliability	15%	Tau-bench, TAU retail/airline, MCP Atlas, Toolathlon, BrowseComp, WideSearch, DeepSearchQA

Important: AGI Readiness is not a claim that AGI has been reached. It is a frontier-mix benchmark proxy: best available model per benchmark. The project also tracks pending benchmarks that need reliable daily sources before being automated.

The current ETA stays in collecting_baseline until the v2 formula has enough fresh daily history. Old history is preserved as legacyHistory because the previous formula is not comparable.

Setup

Frontend

pnpm install
pnpm dev      # dev server
pnpm build    # production build

Data collector

1. Get an API key at llm-stats.com/developer

2. Local test:

LLM_STATS_API_KEY=your_key node scripts/collect.mjs

3. GitHub Actions automation:
   • Repo → Settings → Secrets → Actions → New secret
   • Name: LLM_STATS_API_KEY, Value: your key
   • Workflow runs daily at 02:00 UTC — also triggerable manually

Deploy (GitHub Pages)

• The repo includes .github/workflows/deploy.yml
• Build command: pnpm build
• Output directory: dist
• In GitHub, enable Pages with Source: GitHub Actions
• Every push to master deploys to https://brayanjeshua.github.io/agi-clock/

Contributing

Open an issue to propose new benchmarks (include source, scoring direction, and whether it has a reliable daily API). PRs welcome.

License

MIT — built by Brayan Jeshua