LEADERBOARDS

Separate rankings for each track

P is the total number of model parameters, including all MoE expert parameters and, for vision-language models, vision encoder parameters.

01

TEXT / LLM

Text leaderboard

The Culture Score combines T1 and T2; T3 and T4 are reported separately.

CategoryParameter range
TinyP ≤ 1B
Small1B < P ≤ 8B
Medium8B < P ≤ 30B
LargeP > 30B

02

VISION / VLM

Vision leaderboard

CategoryParameter range
SmallP ≤ 8B
Medium8B < P ≤ 30B
LargeP > 30B