OpenAI Evals
OpenAIFramework for evaluating LLMs and LLM systems plus a registry of community-contributed benchmarks, published as the original 2023 reference implementation for eval-driven development. Picked because it is the canonical starting point referenced in OpenAI's own docs and many derivative harnesses. 18.5k stars but development has effectively slowed (2 commits in last 90 days) as work migrated to the closed OpenAI Dashboard Evals API.
openai/evals GitHub repo (MIT, 18.6k stars, ~691 commits). Open-source framework for evaluating LLMs with a community benchmark registry. Confirmed live on GitHub June 2026; bundled eval datasets carry their own per-dataset licenses (CC/Apache/CC0).
Openness
5 high confidence- license
- MIT(OSI)
- source
- public(full framework + registry)
- per-dataset-licenses
- mixed(CC/CC0/Apache for bundled data)
- core-gated
- ungated
Framework code is fully OSI (MIT); the included eval datasets carry their own open licenses, which is standard for a harness.
- https://github.com/openai/evals/blob/main/LICENSE.md recorded 2026-06-04
MIT License text (copyright 2023 OpenAI)
- https://github.com/openai/evals recorded 2026-06-04
open framework for evaluating LLMs with open benchmark registry; MIT badge; 18.6k stars
Adoption
3 low confidenceNo PyPI/download count available for this repo (it is run from source, not pip-installed); 18.6k GitHub stars + 3k forks is the only honest signal, so capped at level 3 per the stars-fallback rule. Historically influential as one of the first open eval frameworks but largely superseded by the hosted Evals API and by newer harnesses.
- https://github.com/openai/evals recorded 2026-06-04
18.6k stars, 3k forks, 691 commits
Capability
3 medium confidenceSolid general-purpose authoring framework but not frontier on task coverage, backend breadth, or agentic-eval support relative to category leaders.
- https://github.com/openai/evals recorded 2026-06-04
framework + registry of benchmarks, custom eval authoring, model-graded grading
Unchanged since 2026-07-30 (last edited, not re-checked)