OpenCompass
An LLM evaluation platform supporting models from OpenAI, Anthropic, Gemini, Qwen, GLM, DeepSeek and others across more than 100 datasets covering knowledge, reasoning, coding, science and safety.
OpenCompass is a platform for evaluating large language models across more than 100 datasets, covering knowledge, reasoning, coding, science, language, long-context understanding and safety, and supporting a wide range of models from providers including OpenAI, Anthropic, Gemini, Qwen, GLM and DeepSeek.
It integrates with VLMEvalKit for native multimodal dataset loading and evaluation with VLMEvalKit's own metrics, and has expanded its API model ecosystem to support the OpenAI Responses API and the LiteLLM AI Gateway, alongside updated Gemini and Anthropic SDK integrations. Configuration files for datasets, models and summarizers have been consolidated into the main package as of version 0.4.0.
The project publishes CompassHub and CompassRank alongside its documentation and installation instructions.