SynopticBench: Evaluating Vision-Language Models on Generating Weather Forecast Discussions of the Future

Build an evaluation framework for multimodal models that processes chaotic scientific data (weather). This allows researchers to verify model generation quality for physics-heavy tasks.

Suggested repo: synoptic-bench

"Can your VLM predict the weather or just write about it?"

Estimated effort: 50h