Local embedding models
Text · Vectors for semantic search and RAG, and rerankers.
MTEB (Multilingual, v2)
Mean(task) · 0–100 · higher is better
Source: MTEB results dataset (independent runs, CC0) · 15 curated models; other versions and benchmarks are not mixed in
Qwen3-Embedding 8B: 70.6 (Multilingual v2, independent run)
80 models, 13 of them rerankers
28 models under 500M parameters
17 of 19 curated models have Portuguese confirmed by a source
Choose your machine and the charts and tables mark what fits in its memory.
| Qwen3-Embedding 8B | 70.6 | 7.6 | ~12 GB | 4,096 | Apache 2.0 |
|---|---|---|---|---|---|
| Qwen3-Embedding 4B | 69.5 | 4 | ~7 GB | 2,560 | Apache 2.0 |
| Harrier OSS v1 (0.6B) | 69.0 | 0.6 | ~3 GB | 1,024 | MIT |
| Qwen3-Embedding 0.6B | 64.3 | 0.6 | ~3 GB | 1,024 | Apache 2.0 |
| Multilingual E5 Large Instruct | 63.2 | 0.56 | ~3 GB | 1,024 | MIT |
| EmbeddingGemma (300M) | 61.1 | 0.3 | ~3 GB | 768 | Publisher terms: Gemma Terms of Use |
| BGE-M3 | 59.6 | 0.57 | ~3 GB | 1,024 | MIT |
| Nomic Embed Text v2 (MoE) | 57.6 | 0.48 | ~3 GB | 768 | Apache 2.0 |
| Multilingual E5 Base | 57.0 | 0.28 | ~3 GB | 768 | MIT |
| Snowflake Arctic Embed L v2.0 | 57.0 | 0.57 | ~3 GB | 1,024 | Apache 2.0 |
| Granite Embedding 311M Multilingual R2 | 56.0 | 0.31 | ~3 GB | 768 | Apache 2.0 |
| Granite Embedding 278M Multilingual | 53.7 | 0.28 | ~3 GB | 768 | Apache 2.0 |
| BGE Large EN v1.5 | 45.1 | 0.34 | ~3 GB | 1,024 | MIT |
| BGE Base EN v1.5 | 44.5 | 0.11 | ~3 GB | 768 | MIT |
| BGE Small EN v1.5 | 43.8 | 0.033 | ~3 GB | 384 | MIT |
| BGE Multilingual Gemma2 | No sourced MTEB result | 9.2 | ~14 GB | 3,584 | Publisher terms: Gemma Terms of Use |
| Granite Embedding Small English R2 | No sourced MTEB result | 0.048 | ~3 GB | 384 | Apache 2.0 |
| Llama Nemotron Embed 1B v2 | No sourced MTEB result | 1.2 | ~4 GB | 2,048 | Publisher terms: NVIDIA Open Model License Agreement (also Llama 3.2 Community Model License Agreement) |
| Nemotron 3 Embed 1B | No sourced MTEB result | 1.1 | ~4 GB | 2,048 | OpenMDW 1.1 |
| all-MiniLM-L6-v2 | No sourced MTEB result | Not reported, or a community distribution | Unknown parameters: no estimate | Not curated yet | Apache 2.0 |
| all-MiniLM-L6-v2-4bit | No sourced MTEB result | Not reported, or a community distribution | Unknown parameters: no estimate | Not curated yet | Apache 2.0 |
| all-mpnet-base-v2 | No sourced MTEB result | 0.1 | ~3 GB | Not curated yet | Apache 2.0 |
| bge-base-en | No sourced MTEB result | 0.1 | ~3 GB | Not curated yet | MIT |
| bge-base-zh | No sourced MTEB result | 0.1 | ~3 GB | Not curated yet | MIT |