Home › Benchmarks › HumanEval & GSM8K
HumanEval & GSM8K
Benchmark
Capability
The retired classics of coding and grade-school math.
You're looking at a benchmark record: a test the industry uses to measure chips or models — and what it actually tells you.
HumanEval (164 Python functions) and GSM8K (grade-school word problems) measured the 2021–24 climb from barely-working to near-perfect, then saturated — their history is the cleanest record of how fast base capability moved. Successors: LiveCodeBench, SWE-bench, AIME.
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Related
Sources
Record reviewed as of 2026-01-15 — see Methodology.