Benchmark explainers
What each benchmark actually measures
The benchmarks everyone argues about, explained from the record — what they measure, how they're scored, and how much to trust them.
integrity grade B
What is SWE-bench Verified?
32 models scored · top score 83.47
integrity grade C
What is MMLU?
99 models scored · top score 84.13
integrity grade B
What is GPQA diamond?
153 models scored · top score 93.1
integrity grade A
What is ARC-AGI-2?
71 models scored · top score 92.5
integrity grade B
What is ARC-AGI?
72 models scored · top score 98.5
integrity grade A
What is AIME?
8 models scored · top score 79.8
integrity grade A
What is FrontierMath Tier 4?
48 models scored · top score 87.8
integrity grade A
What is SimpleQA Verified?
68 models scored · top score 77.3
integrity grade A
What is Chess Puzzles?
81 models scored · top score 62.12
integrity grade A
What is LiveCodeBench?
8 models scored · top score 65.9
integrity grade A
What is HLE?
37 models scored · top score 43.74
integrity grade C
What is GSM8K?
56 models scored · top score 92