AI Potluck
Model components / Evaluation code

OpenCompass

OpenCompass Community

Evaluation platform covering more than a hundred datasets across open and API models, with both objective scoring and subjective arena-style comparison. Backed by Shanghai AI Lab, it has notably deeper Chinese-language and multimodal coverage than the English-first harnesses.

Verified 2026-08-13 via the open-compass/opencompass repository and its license endpoint.

Openness

5 high confidence
5.0
license
Apache-2.0(OSI)
source
public(GitHub)
core-gated
ungated

Apache-2.0 and fully public, and Apache-2.0 is an OSI license, so this is open source.

Adoption

3 medium confidence
3.0

A widely used eval framework, especially in the China and Hugging Face ecosystems, with 7.1k GitHub stars and broad dataset and backend coverage. No clean download headline is available, so the level rests on reported traction - framework distribution plus the star footprint - rather than on stars alone, which would cap it lower. A PyPI package named opencompass does exist and is genuinely this project's, but it is not the measure used: it sees 5,728 downloads a month against a project whose documented path is a git clone and a config tree, so banding on it would read a minority channel rather than the project.

Capability

4 medium confidence
4.0

Very broad multi-benchmark coverage and backend breadth, sitting just below lm-eval as the standardization reference - strong, but not the single de-facto global standard. The README claims 100+ datasets and lists Hugging Face, LMDeploy, vLLM and API backends as install extras.

Verified 2026-08-13