Benchmarks
Every benchmark the index tracks, by tier, with each core benchmark's share of the intelligence index and how many models it covers.
700 benchmarks tracked; 10 weigh in the intelligence index.
- AIME (annual) · 187 models
- ARC-AGI-2 · 74 models
- LMArena / Arena Elo · 170 models
- GPQA Diamond · 322 models
- Humanity's Last Exam · 280 models
- LiveBench · 90 models
- MMLU-Pro · 188 models
- SWE-bench Verified · 130 models
- tau-bench / tau2-bench · 196 models
- Terminal-Bench (Hard) · 270 models