← All notes

Summary

Holistic Evaluation of Language Models (HELM) is a multi-metric, multi-scenario evaluation framework from Stanford CRFM. The authors measure 7 metrics on 16 core scenarios when possible (they report 87.5% of the time), and they evaluate 30 language models on 42 scenarios. They say that before HELM, models were evaluated on average on just 17.9% of the core scenarios, and that their work raises this to 96.0% under standardized conditions. They report 25 top-level findings and release prompts, completions, and a modular toolkit. [S1]

Claims

• id: C1 claim: 'They measure 7 metrics for each of 16 core scenarios when possible.' source: S1 evidence: "for each of 16 core scenarios when possible"
• id: C2 claim: 'They evaluate 30 prominent language models on all 42 scenarios.' source: S1 evidence: "a large-scale evaluation of 30 prominent language models"
• id: C3 claim: 'They report improving coverage of core scenarios from 17.9% to 96.0%.' source: S1 evidence: "evaluated on just 17.9% of the core HELM scenarios"

Verification

• Code available: Paper: https://github.com/stanford-crfm/helm (HTTP 200 on 2026-11-11) and project page https://crfm.stanford.edu/helm/v1.0 (HTTP 200). Not run.
• Reproduced: not reproduced (2026-11-11). Code not run.
• Related work: Live note 2309-15217-ragas-rag-evaluation. This is the paper-verification note for HELM itself.
• Limits: HELM is a living benchmark; the v2 paper (revised 1 Oct 2023) describes an early public snapshot, including the v1.0 project page named in the comments, and later site versions may add scenarios or models.
• License: CC BY 4.0 on the abs page (2026-11-11; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 16 Nov 2022 (v1); last revised 1 Oct 2023 (v2). Sources: [S1] https://arxiv.org/abs/2211.09110, v2 (1 Oct 2023). checked_on: 2026-11-11.

When to apply

Use when you need a multi-metric evaluation frame, or when accuracy-only score tables hide trade-offs.

Facts

• id: F1 claim: 'HELM measures 7 metrics on 16 core scenarios when possible, reported as 87.5% of the time.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F2 claim: 'The authors evaluate 30 language models on 42 scenarios.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F3 claim: 'They report that average coverage of core HELM scenarios rose from 17.9% before HELM to 96.0% in their evaluation.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0

Sources

• [S1] arXiv 2211.09110, https://arxiv.org/abs/2211.09110. License: CC BY 4.0. v2 (1 Oct 2023). Accessed 2026-11-11.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.

@misc{iil_note_6qgeh8x6,
  author = {{'iinventedlight (AI-assisted)'}},
  title = {{Holistic Evaluation of Language Models}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-11},
  url = {https://iinventedlight.com/library/notes/2211-09110-helm/},
  note = {Permanent ID iil:note:6qgeh8x6. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2211-09110-helm.md.