ASIndex tracks publicly accessible AI systems from today's frontier capabilities toward artificial superintelligence on a practical 0-100 scale. Scores reflect demonstrated, broad, real-world capability, not isolated benchmarks, private laboratory demonstrations, or narrow specialties.

Each system is evaluated at its highest generally available reasoning setting under standardized conditions using a private, offline benchmark suite designed to minimize contamination. ASIndex retests active frontier-class models every Friday, including models whose public version name has not changed, because hosted AI systems can improve through updates to weights, routing, tools, system behavior, or serving infrastructure. Newly available frontier-class models are tested as soon as practical. Eligibility requires access through a broadly available product, API, platform, or comparable public channel under standard terms. Systems that depend on restricted access, bespoke private deployments, or non-public infrastructure are not eligible for the active index. When a tested system is withdrawn from public availability, it moves to the historical archive and is removed from the current leaderboard.

Read the full ASIndex definition and testing rules.