Back to Gap Map Model components / Evaluation code

OpenCompass

OpenCompass Community
open source / Overall score: 3.5

Evaluation platform covering more than a hundred datasets across open and API models, with both objective scoring and subjective arena-style comparison. Its dataset list includes many Chinese-language benchmarks, and multimodal evaluation runs through the integrated VLMEvalKit. The project is developed at Shanghai AI Laboratory.

Openness

5 high confidence
5.0
license
Apache-2.0(OSI)
source
public(GitHub)
core features withheld
no

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

  • https://api.github.com/repos/open-compass/opencompass/license recorded 2026-08-13

    GitHub's license endpoint reports spdx_id Apache-2.0 for open-compass/opencompass and returns the Apache-2.0 text as the LICENSE body.

  • https://github.com/open-compass/opencompass recorded 2026-06-04

    Apache-2.0 license, public source, v0.5.2 (Feb 2026)

  • https://pypi.org/pypi/opencompass/json recorded 2026-09-29

    PyPI metadata for opencompass 0.5.4 lists the license as "Apache License 2.0"; the packaged README offers a source install via "git clone https://github.com/open-compass/opencompass opencompass" and a plain "pip install -U opencompass" with no enterprise, pricing, paid tier or license key wording anywhere in the body.

  • https://raw.githubusercontent.com/open-compass/opencompass/main/README.md recorded 2026-08-13

    README installs the platform with `pip install -U opencompass` or editable from a clone and runs evaluations locally against local weights or model APIs; all dataset and model configs ship inside the package. CompassRank and CompassHub are free public leaderboard and dataset sites beside it. No pricing, enterprise edition, paid tier or license key appears in the README.

Adoption

3 medium confidence
3.0

OpenCompass is widely used in the China and Hugging Face ecosystems, with broad dataset and backend coverage and a GitHub star footprint to match. A PyPI package named opencompass is genuinely this project's, but its documented install path is a git clone, so banding on its downloads would read a minority channel rather than the project, leaving reported framework use as the basis instead.

Capability

4 medium confidence
4.0

OpenCompass documents well over a hundred supported datasets and runs models through Hugging Face, LMDeploy, vLLM or hosted APIs, with both objective and subjective scoring. It sits a step below lm-evaluation-harness, which has been used to report results in hundreds of papers and is the more established reference.

Verified 2026-09-29, except some scores, still awaiting re-confirmation