Safety is a capability dimension inside a general evaluation framework, measured alongside reasoning and knowledge rather than treated as a separate compliance track.
Evaluation PlatformChina
OpenCompass at Shanghai AI Laboratory
OpenCompass 司南
Shanghai AI Laboratory's open evaluation system is one of the few public windows into how a Chinese frontier lab measures model capability, and it carries safety as an explicit dimension.
Importance
Assessed per aspect, band 0 to 3. There is no composite score.
Regulatory authoritynot assessed
Standards influence
1 · emerging
Its scores circulate as public reference points; there is no recorded adoption of its metrics into formal requirements.
analyst judgmentsrc-opencompass-doc
Technical capacity
2 · established
It runs an open, reproducible evaluation framework that has expanded into safety and trustworthiness dimensions.
Convening powernot assessed
International interfacenot assessed
What safety means here
Leadership
No leadership roles are recorded with a verifiable source.
History
- 2023-10-19
OpenCompass launched as an open, reproducible evaluation framework.
- 2025-05-16
Scientific intelligence evaluation system launched under OpenCompass.
Connections
One recorded edge: Shanghai AI Laboratory maintains the platform. analyst judgment
Maintains
- Shanghai Artificial Intelligence Laboratory SHLAB's official news page describes OpenCompass as built by SHLAB and used as a broad evaluation system.
Selected outputs
Engagement fit
Assessmentanalyst judgment
- Plausible for
- Technical exchange on evaluation methods and benchmark design, in the open-source register.
- Constrained by
- A platform, not an institution; positions and priorities belong to the lab that maintains it.
Open questions
- How its safety dimension relates to the lab's standards work in WG9, which the current sources do not connect.
Source appendix (4 records)
- src-opencompass-launch
Shanghai AI Laboratory launches OpenCompass · Shanghai AI Laboratory · 2023-10-19open
- src-opencompass-doc
OpenCompass framework overview · OpenCompass · unknownopen
- shlab-opencompass-2023
OpenCompass Launches Scientific Intelligence Evaluation System · Shanghai Artificial Intelligence Laboratory · 2025-05-16open
- report-technical-research-groups-2026-06-05
Technical Research Groups Research Brief · Asia AI Safety Atlas research brief
Last verified 2026-07-07