HumanEval

HumanEval: hand-written Python programming problems (Codex paper)

164 hand-written Python programming problems, each a function signature plus docstring, scored by running hidden unit tests and reporting pass@k (usually pass@1) for functional correctness. It introduced execution-based code evaluation and was the standard code benchmark from 2021 to 2024; frontier models now solve nearly every problem and the set is widely present in training data.

99.3%
o4-mini-high
self-reported
2021-07-07 → 2026-04-22: 28.8% → 76.8%
Released
2021-07
Maintainer
OpenAI
Status
saturated
Contamination
high
Metric
pass@1 (percent, ↑)
Tasks
164
Domains
code
human
no measured baseline

Notes. Many papers report EvalPlus (HumanEval+) numbers with extra tests, which are lower and not directly comparable. OpenAI's simple-evals stopped reporting new models in July 2025.

Full ledger

SystemDeveloperScoreDateSourceConditions
o4-mini-highOpenAI99.3%self-reportedpass_k: 1 shots: 0
simple-evals README; effectively the ceiling (163 of 164).
gpt-4-turbo-2024-04-09OpenAI88.2%self-reportedpass_k: 1 shots: 0
simple-evals README.
DeepSeek-V4-Pro-BaseDeepSeek76.8%self-reportedpass_k: 1 shots: 0
DeepSeek-V4 model card, base-model table (HF repo created 2026-04-22). Base (pre-trained, not instruction-tuned) model, so lower than chat models above.
Codex 12BOpenAI28.8%paperpass_k: 1
Codex paper abstract; Codex-S 37.7, GPT-J 11.4, GPT-3 0. 70.2% with 100 samples per problem.