JevBench, a reproducible benchmark for typed decision models
5 hours ago
- The table ranks 48 models across multiple performance metrics (columns #1 to #5).
- Jev 1.13.0 is the top-ranked model overall, achieving a score of 74.4 in the first column.
- Many models show ranks that differ from their JevBench Score, indicating variability in performance across different tasks.
- The lowest performing models, such as Certo v1, Alibaba GTE Reranker ModernBERT-base, and BAAI bge-reranker-v2-m3, have near-zero scores.
- The models vary widely in size and type, from small models like kev 0.5B to larger ones like SimpleJev Qwen3.6-35B-A3B.