Vision-Language Models and Benchmarking

RocketScience

Contrastive benchmark for spatial reasoning in VLMs using real-world data.

TopoBench

Synthetic benchmark for super long horizon spatial reasoning on board game puzzles with verifier based grading.

Implemented RocketScience in openbench

Merged into main.

FrontierBenchmarks

Interactive live benchmark map for tracking the changing frontier of language model evaluations.