FlagEval

SCIENTIFIC

Beijing Academy of Artificial Intelligence (BAAI) is one of the earliest scientific research institutions in the world to systematically layout large models. Based on profound technical accumulation, it has launched the "BAAI Evaluation System", which examines the cognitive ability of large models from multiple dimensions, covering large models in multiple fields such as natural language processing, computer vision, and multimodality. It conducts comprehensive and systematic evaluations on global open source and commercial closed source models.

FAIR

As a non-profit research institution, BAAI has a rare neutral perspective in the industry. In terms of evaluation methods, objective evaluation ensures that the evaluation is conducted in a unified environment, standardizing the "starting line"; subjective evaluation sets frequently iterate and strictly calibrate scoring standards, and a large number of original and non-public evaluation sets are constructed to prevent data leakage.

AUTHORITATIVE

BAAI participates in and leads the formulation of domestic and international standards related to large model evaluation, striving to promote the consensus of "large model evaluation methods and norms" in the industry; and takes the lead in cooperating with authoritative education departments to horizontally align the cognitive level that large models can achieve based on the development ladder of human cognitive ability.

OPEN

BAAI is cooperating with more than ten research institutions to carry out the construction of evaluation datasets and evaluation tools. In the future, we will uphold the spirit of "open source and openness", take "evaluation" as the breakthrough point, connect more large model R & D teams, evaluation research groups, and large model industry application enterprises, and jointly promote the optimization of large model technology, application landing, and ecological sharing.

FLAGEVAL

SCIENTIFIC

FAIR

AUTHORITATIVE

OPEN