Opencompass

OpenCompass is an open, comprehensive evaluation platform designed for developers and users of large AI models. It focuses on providing standardized, transparent, and reproducible assessments for large language models (LLMs) and multimodal models. The platform aggregates benchmark results and evaluation resources to help users understand model capabilities across a wide range of tasks and domains.

The platform offers curated leaderboards for both text-based and multimodal models, presenting multi-dimensional scores across capabilities such as reasoning, understanding, generation quality, and robustness. Its benchmark suite spans general knowledge, coding, math, instruction following, safety, and more, enabling fine-grained comparison of models in realistic scenarios. OpenCompass also hosts an evaluation set community, where researchers and practitioners can contribute, share, and discover innovative benchmark datasets, along with detailed metadata and documentation. Comprehensive documentation and tooling support make it easier to reproduce evaluations and integrate new models into the framework.