Eventual-Inc/VLADBench-reeval
VLADBench-reeval We re-evaluated the VLADBench benchmark (Li et al., 2025, arXiv:2503.21505) against current SOTA VLMs under the original scoring criteria and prompts. See Eventual-Inc/VLADBench for the code, the companion site for interactive results, and the article for a summary of the findings. Cost vs Score Up and to the left is better. The dashed line is the cost-performance frontier which is defined by measuring the best TOTAL score available against its… See the full description on the dataset page: https://huggingface.co/datasets/Eventual-Inc/VLADBench-reeval.
Results as of 2026-09-23T04:13:26Z: 15 models, 167,895 answers
Card: link the article; full author list in the citation
Results as of 2026-09-23T04:13:26Z: 15 models, 167,895 answers
Results as of 2026-09-22T20:31:36Z: 13 models, 145,509 answers
Results as of 2026-09-22T20:31:36Z: 13 models, 145,509 answers
Results as of 2026-09-16T04:34:41Z: 12 models, 134,316 answers
Results as of 2026-09-16T04:34:41Z: 12 models, 134,316 answers
Results as of 2026-09-16T04:34:41Z: 12 models, 134,316 answers
Results as of 2026-09-16T04:34:41Z: 12 models, 134,316 answers
Results as of 2026-09-16T04:34:41Z: 12 models, 134,316 answers
Results as of 2026-09-16T04:34:41Z: 12 models, 134,316 answers
Results as of 2026-09-16T04:34:41Z: 12 models, 134,316 answers
initial commit