S1MB benchmark uses Borda ranking to highlight consistent AI model performance

The System One Mosaic Benchmark evaluates 102 AI models across 137 specialized tasks in three categories. It employs a Borda scoring method, which ranks performance on each task and sums the ranks to emphasize broad capability over excelling in only a few areas. The current leader is the Darwin-27B-ZTC-v2 model, which holds a narrow lead over close competitors. This ranking approach is designed to make the top position more meaningful by preventing models from gaming the system through overfitting. The benchmark uses a zero-token judging method that ensures reproducible and exact scores without sampling variance.