Video is the widest gap: WeMM-Embedding-2B beats an 8B by 3.7 points
tomaarsen · x · 2026-09-02
The video subset of MMEB-v2 is where WeMM-Embedding leads by the widest margin:
- WeMM-Embedding-9B: 74.3
- WeMM-Embedding-2B: 70.8
- Qwen3-VL-Embedding-8B: 67.1
The 2B model beats the 8B Qwen3-VL embedding by 3.7 points on video, making video retrieval the family's standout capability.
More from Multimodal
- Flux 3 Text-to-Image Can't Make Realistic Faces at All, Reddit User Reports — EstablishmentFun6629 · 2026-10-18
- MiniMax H3 Ref-to-Video: 15s Shot from 3 Images + Vocals, 1088p Latent Refinement — michael_goldenhour · 2026-10-18
- Dev builds fully local AI tool generating HD 3D textures with Qwen, may sell for $5 — TinfoilTricorn · 2026-10-18
- 14-Agent AI Film Pipeline on One RTX 5090: Fixing Cut Defects Between Clips — developervkmp · 2026-10-18
- 4MP Text-to-Image Comparison: Qwen Image 2.1 Turbo vs MOSS vs Krea 2 vs Flux 3 Klein — FORNAX_460 · 2026-10-18
- Claude writes entire song from just lyrics: "Don't let me scroll" — TinfoilTricorn · 2026-10-18