# openai/human-eval

Code for the paper "Evaluating Large Language Models Trained on Code"

Repository: https://github.com/openai/human-eval
Canonical: https://ross.abutalabs.com/products/human-eval
Language: Python
License: MIT
License Family: permissive
Last push: 2025-01-17T18:22:17+00:00

## Health v2 (maintenance only)
Score: 33/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 2, release rhythm 35, longevity 100
- inputs: {"age_days": 1884, "days_push": 593, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3354, forks 458 (observed 2026-08-28T04:07:57.872852+00:00)

## What it is
HumanEval is a hand-written evaluation dataset of 164 Python programming problems with unit tests, plus a harness for measuring code-generation model performance via pass@k functional correctness. It accompanies OpenAI's paper 'Evaluating Large Language Models Trained on Code'.

## Use cases
- benchmark a code-generation LLM on HumanEval
- compute pass@1 pass@10 pass@100 for code completions
- evaluate functional correctness of generated Python code
- compare code models against a standard coding benchmark
- run unit tests against model-generated completions

## When to choose
- you need a standard, widely-cited benchmark for code generation models
- you want functional-correctness scoring with unit tests rather than string matching
- you are reproducing results from code LLM papers

## When to avoid
- you need a sandboxed execution environment - the harness runs untrusted generated code and deliberately leaves sandboxing to you
- you need evaluation in languages other than Python
- you need a maintained multi-task benchmark suite - this is a single frozen dataset

## Facets
- artifact type: dataset
- maturity: maintenance
- function: testing, benchmarking
- domain: large-language-models, machine-learning, developer-tools, programming-languages
- platform: python, cli, cross-platform
- tags: code-generation, llm-evaluation, humaneval, pass-at-k, benchmark-harness, unit-tests, evaluation

## Member repositories
- openai/human-eval (main) score 33

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:57.872852+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:40:56.279527+00:00, confidence not recorded.
  - readme: https://github.com/openai/human-eval (fetched 2026-08-28T04:07:57.872852+00:00, sha 607df2462951)
- Data as of 2026-08-30T08:39:29.467469+00:00.