A transparent, reproducible evaluation of general language-model capabilities across a curated collection of difficult scenarios.
These scores are reproduced from the linked primary source. They have not been re-evaluated or adjusted by Shareof.ai.
Stanford CRFM evaluates models on a versioned collection of capability scenarios and publishes prompts, responses, metrics and reproducible configurations.
Reported metric: Mean score. A higher score represents stronger measured performance within this benchmark.
Scores apply only to HELM Capabilities version 1.14.0 and should not be compared with scores from a different benchmark version.
Use the leaderboard to compare configurations evaluated under this exact methodology. Do not compare these values directly with a similarly named metric from another benchmark.
A standardized collection of safety evaluations spanning violence, fraud, discrimination, harassment, sexual content and deception.
A reproducible evaluation of how reliably models reason over long documents, conversations and multi-hop evidence.
A healthcare-focused evaluation framework grounded in real clinical tasks, including reasoning and patient communication.
A difficult set of open-ended real-world prompts evaluated with automated judges and style controls.
Run a live visibility check and compare your brand with the companies appearing in the same AI answers.
Benchmark my business →