Every figure on these pages links to its source, every update is dated, and every change is logged, so you can check our work and cite it.
Observatory
AI Agent Benchmark Observatory
What eight agent benchmarks measure, how they are graded, what runs cost, who leads each maintainer's own leaderboard, and when the methods changed.
- Covers
- GDPval · SWE-bench · Terminal-Bench · OSWorld · RLI · METR
- Updated
- Snapshots refreshed by hand, each with a dated changelog entry
- Formats
- Web · CSV · JSON
Live leaderboard
ARC-AGI-3 leaderboard
The best published score for each frontier model on ARC Prize's semi-private set: the share of unfamiliar interactive games each one solves.
- Covers
- ARC-AGI-3 · ARC Prize 2026
- Updated
- Checked against ARC Prize's own data every few hours
- Formats
- Web
Tracker
Linux patch status tracker
Whether each major Linux distribution has shipped a fix for high-profile vulnerabilities, with the fixed package and the vendor advisory.
- Covers
- CVE-2026-98432 · PHP2Shell
- Updated
- Updated when a distribution ships a fix, every change logged
- Formats
- Web
Found an error, or know a dataset we should track? Email [email protected].