LiveCodeBench Team

data

downloadable(HF parquet)

license

cc(the card YAML names the Creative Commons family and no variant

datasheet

present(README documents versions/sources/structure)

redistributable

yes

splits

public problems with hidden test cases bundled

LiveCodeBench's card names only the Creative Commons family, "cc", with no stated variant anywhere the project publishes - the repository, README, site or paper - and its MIT license covers only the lcb_runner harness, not the problem corpus. Files download freely, but without a stated permission to use them, the corpus reads as restricted rather than open.

• https://huggingface.co/api/datasets/livecodebench/code_generation_lite recorded 2026-09-29

  The API JSON lists the data files "test.jsonl" and "test6.jsonl" for direct download and tags the repo "license:cc" with no variant; the description reads "LiveCodeBench is a temporaly updating benchmark for code generation". Searched the body for noncommercial and by-nc: none.

• https://huggingface.co/datasets/livecodebench/code_generation_lite recorded 2026-09-29

  The dataset page tags the repo "license:cc" with no variant and renders the card with "Each problem instance consists of a problem description, input/output examples, and hidden test cases". Searched the body for non-commercial, by-nc and holdout terms: none.

• https://huggingface.co/datasets/livecodebench/code_generation_lite/blob/main/README.md recorded 2026-09-29

  The blob view of the card renders "Since LiveCodeBench is a continuously updated benchmark" and its release change log. Searched the body for by-nc, noncommercial and holdout terms: none.

• https://huggingface.co/datasets/livecodebench/code_generation_lite/raw/main/README.md recorded 2026-09-29

  The card front matter reads "license: cc" with tags, pretty_name and size_categories, naming no Creative Commons variant; the body carries a change log ("release_v5" with 880 problems) and the statement "Each problem instance consists of a problem description, input/output examples, and hidden test cases."

• https://raw.githubusercontent.com/LiveCodeBench/LiveCodeBench/main/LICENSE recorded 2026-08-13

  MIT License, copyright 2024 LiveCodeBench. This is the repository holding lcb_runner, the evaluation harness; the problems are distributed from the Hugging Face repos and this file makes no statement about them

• https://raw.githubusercontent.com/LiveCodeBench/LiveCodeBench/main/README.md recorded 2026-08-13

  217 lines covering install, running the harness, scenarios and submissions, with no licensing or terms section at all

Adoption is measured by Hugging Face downloads of livecodebench/code_generation_lite, the benchmark's only declared distribution artifact.

• https://huggingface.co/api/datasets/livecodebench/code_generation_lite recorded 2026-08-13

  91,719 downloads in the trailing 30 days for livecodebench/code_generation_lite

A dataset is not 'capable', so capability is left unscored. Its contamination resistance - the "live", continuously updated design - is a real quality strength, but openness and adoption are what carry this category.

• https://huggingface.co/datasets/livecodebench/code_generation_lite recorded 2026-09-29

  The page describes a problem corpus: "Each problem instance consists of a problem description, input/output examples, and hidden test cases", released in versions; nothing on it claims a capability of its own. Searched the body for throughput, accuracy, SOTA and pass@1: none.