HELM vs LiveBench
HELM
crfm.stanford.eduLiveBench
livebench.aiSide by side
| Figure | HELM | LiveBench |
|---|---|---|
| Price from | Free | — |
| Free plan | Yes | — |
| Free trial | — | — |
| Open source | — | — |
| Web | Yes | Yes |
| Windows | — | Yes |
| macOS | — | Yes |
| Linux | — | Yes |
| API | Yes | — |
| Self-hosted | Yes | Yes |
| Plans published | 1 | 0 |
| Spec score · LLM Evaluation Tools | 8.3 (#7) | 7.2 (#22) |
| Purpose | HELM is an open-source Python framework for holistic, reproducible, and transparent evaluation of foundation models, including language and multimodal models. | LiveBench is an LLM benchmark designed to limit test set contamination and use objective evaluation. |
| Support | Users are directed to open GitHub issues for questions, which maintainers address when bandwidth is available. | The repository invites model evaluation requests by GitHub issue or email at [email protected]. |
Plans
In their own words
HELM
HELM is an open-source Python framework for evaluating foundation models, including large language and multimodal systems. It supplies standardized datasets and benchmarks such as MMLU-Pro, GPQA, IFEval, and WildBench, with a unified interface for models from providers including OpenAI, Anthropic, and Google. Evaluations can examine accuracy, efficiency, bias, and toxicity, and can include custom metrics, LLM-as-a-judge checks, safety evaluations, and human review workflows. Users can configure private checkpoints or model servers, run inference locally, or send requests to an API. A web interface displays individual prompts and responses, while leaderboards compare models and benchmarks across areas such as medicine, finance, multilinguality, and world knowledge. HELM can reproduce published evaluation results and documents commands for running benchmarks, summarizing results, and starting a local server. It is free, Apache-2.0 licensed, and entered maintenance mode on June 1, 2026.
LiveBench
LiveBench is an open-source benchmark for evaluating large language models, designed to limit test-set contamination and use objective scoring. Its current scope is 23 tasks across Reasoning, Coding, Agentic Coding, Mathematics, Data Analysis, Language, and Instruction Following, with a refresh every six months. The leaderboard presents overall and category scores, and model rows expand to show subtasks. Insight views cover quality versus cost, ranked cost, and category profiles; cost per successful task is calculated as (Σ cost ÷ Σ questions ÷ score) × 100 for the selected scope. Questions have verifiable ground-truth answers and can be scored automatically without an LLM judge. The public project provides benchmark code and links to data on Hugging Face, as well as a Python command-line evaluation pipeline. It supports OpenAI-compatible endpoints and implemented inference for Anthropic, Cohere, Mistral, Together, and Google models. Local inference is unmaintained. Agentic coding evaluations require Docker, and task images may take up to 150 GB to store.