Benchmarks and Their Limits
What a benchmark actually measures, how benchmarks saturate and get gamed, and how to read a leaderboard number without being fooled.
Content last verified 2026-09.
Lessons
- What a Benchmark Is
- Saturation: When the Ceiling Arrives
- Gaming, Goodhart, and Leaderboards
- Reading a Benchmark Number
Sources
- Hendrycks et al. (2020) — Measuring Massive Multitask Language Understanding (MMLU)
- Chen et al. (2021) — Evaluating Large Language Models Trained on Code (HumanEval, pass@k)
- Cobbe et al. (2021) — Training Verifiers to Solve Math Word Problems (GSM8K)
- Liang et al. (2022) — Holistic Evaluation of Language Models (HELM)