WeMM-Embedding tops MMEB-v3: 9B scores 59.5, 2B beats every 7B/8B model
tomaarsen · x · 2026-09-02
On MMEB-v3 (190 tasks covering text, agent, and cross-modal retrieval), the WeMM-Embedding family scores:
- WeMM-Embedding-9B: 59.5, top of the board
- WeMM-Embedding-4B: 58.2
- WeMM-Embedding-2B: 56.0
The 2B variant already beats every 7B and 8B model listed.
The thread also shows a SentenceTransformer usage snippet with trustremotecode=True, encoding queries as text and documents as either text or {"video": ...} items.
Related event: Tencent Open-Sources WeMM-Embedding, Tops MMEB Benchmarks(6 posts)→
More from Multimodal
- Sprite Fusion generates production-ready AI pixel art at native sizes in seconds — HugoDuprez · 2026-09-02
- Cyberpunk alien gangster casting video makes the rounds on Reddit — LuckyFeelingPunk · 2026-09-02
- Seedance 2.5 renders insane Shibuya Marvel crossover video — heypearlai · 2026-09-02
- Fable 5.1 + Blender WIP dragon lair showcase — majidmanzarpour · 2026-09-02
- One ChatGPT prompt that fakes iPhone 16 Pro Max photos to build ultra-realistic AI influencers — EXM7777 · 2026-09-02
- Reusable Nano Banana prompt turns any brand logo into photoreal macarons — azed_ai · 2026-09-02