Curated lists / Harness
Benchmark harness: Best Embedding Models for Multimodal Search
Everything needed to reproduce the numbers: the dataset builder, the provider runner, the scorer, and the raw per-file output each provider returned. API keys are read from environment variables at run time and are not included.
Used by: Best Embedding Models for Multimodal Search
- build_dataset.py2,676 B
- results/2026-09-10/clip-vit-b-32.json218 B
- results/2026-09-10/clip-vit-l-14.json219 B
- results/2026-09-10/cohere-embed-v4.json225 B
- results/2026-09-10/jina-clip-v2.json257 B
- results/2026-09-10/nova-multimodal.json249 B
- results/2026-09-10/queries.json229,262 B
- results/2026-09-10/scores.json3,426 B
- results/2026-09-10/siglip2-base.json238 B
- results/2026-09-10/siglip2-so400m.json244 B
- results/2026-09-10/titan-multimodal.json236 B
- run.py12,495 B
- score.py1,974 B