Benchmark

An AI benchmark is a standardized task or collection of tasks used to compare the performance of AI systems.

Benchmarks make progress measurable, but they are imperfect proxies for intelligence. A model can perform extremely well on a benchmark without demonstrating the broader adaptability required in the real world.