NV-Embed-v2
NVIDIA's MTEB champion — 7B LLM-embedder with latent attention, but CC-BY-NC blocks commercial use.
About
NV-Embed-v2 (2024, NVIDIA) was the MTEB champion at release: a 7B Mistral-based embedder with latent-attention pooling that topped the leaderboard and held it. 142k Hugging Face downloads.
Licensed CC-BY-NC-4.0: NON-COMMERCIAL. The critical caveat — NVIDIA's top embedder cannot be used in commercial products as weights. (Commercial access via NVIDIA NIM.) The latent-attention architecture was a genuine advance over mean-pooling. A 2024 model; the MTEB crown has since moved on, but the architecture remains influential. Research use or NVIDIA NIM for products.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-claimed at release (2024): #1 on MTEB — treat as provider-claimed and dated; the crown has moved on. Architecture remains influential.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier). The 2024 MTEB champion with a genuinely novel latent-attention architecture and 32K context. Docked four points: (1) CC-BY-NC-4.0 — no commercial use of weights; (2) the MTEB crown has moved on; (3) 7B is GPU-hungry; (4) modest adoption (142k downloads). Research use or NIM for products.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.2/10
- Maintenance · recency, releases, issue load
- 4.0/10
- Openness · license + docs
- 4.2/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · HF downloads 142,302 (peer-normalized)
- · released 872 days ago
- · manual review: 2024 MTEB champion, latent-attention architecture; NC license; 7B GPU-hungry; license: CC-BY-NC-4.0 — NON-COMMERCIAL; commercial via NVIDIA NIM