MTEB Multilingual v2 leaderboard: who is #1
MTEB · Mean task score, higher is better · 96 models · published Nov 8, 2026
Top 25 of 96 models, best variant per model as published
| # | Model | Mean task score | Score | Correct |
|---|
| 1 |  harrier-oss-v1-27b (microsoft)as “microsoft/harrier-oss-v1-27b” | | 74.3% | |
| 2 |  KaLM-Embedding-Gemma3-12B-2511 (tencent)as “tencent/KaLM-Embedding-Gemma3-12B-2511” | | 72.3% | |
| 3 |  llama-embed-nemotron-8b (nvidia)as “nvidia/llama-embed-nemotron-8b” | | 69.5% | |
| 4 |  Qwen3-Embedding-8B (Qwen)as “Qwen/Qwen3-Embedding-8B” | | 70.6% | |
| 5 |  gemini-embedding-001 (google)as “google/gemini-embedding-001” | | 68.4% | |
| 6 | Qwen3-Embedding-4B (Qwen)as “Qwen/Qwen3-Embedding-4B” | | 69.5% | |
| 7 | OOcten-Embedding-8B (Octen)as “Octen/Octen-Embedding-8B” | | 67.8% | |
| 8 | CAF2LLM-v2-14B (codefuse-ai)as “codefuse-ai/F2LLM-v2-14B” | | 68.7% | |
| 9 | CAF2LLM-v2-8B (codefuse-ai)as “codefuse-ai/F2LLM-v2-8B” | | 68.1% | |
| 10 | harrier-oss-v1-0.6b (microsoft)as “microsoft/harrier-oss-v1-0.6b” | | 69.0% | |
| 11 |  Seed1.6-embedding-1215 (Bytedance)as “Bytedance/Seed1.6-embedding-1215” | | 70.3% | |
| 12 | CAF2LLM-v2-4B (codefuse-ai)as “codefuse-ai/F2LLM-v2-4B” | | 67.1% | |
| 13 | ASGiga-Embeddings-instruct-10B-A1.8B-0826 (ai-sage)as “ai-sage/Giga-Embeddings-instruct-10B-A1.8B-0826” | | 65.6% | |
| 14 | Jjina-embeddings-v5-omni-small (jinaai)as “jinaai/jina-embeddings-v5-omni-small” | | 67.0% | |
| 151up from #16 since Oct 2, 2026 | CAF2LLM-v2-1.7B (codefuse-ai)as “codefuse-ai/F2LLM-v2-1.7B” | | 65.2% | |
| 161up from #17 since Oct 2, 2026 | BBidirLM-Omni-2.5B-Embedding (BidirLM)as “BidirLM/BidirLM-Omni-2.5B-Embedding” | | 63.5% | |
| 171up from #18 since Oct 2, 2026 | ASGiga-Embeddings-instruct-3B-0826 (ai-sage)as “ai-sage/Giga-Embeddings-instruct-3B-0826” | | 63.9% | |
| 181up from #19 since Oct 2, 2026 | harrier-oss-v1-270m (microsoft)as “microsoft/harrier-oss-v1-270m” | | 66.5% | |
| 191up from #20 since Oct 2, 2026 | Qwen3-Embedding-0.6B (Qwen)as “Qwen/Qwen3-Embedding-0.6B” | | 64.3% | |
| 201up from #21 since Oct 2, 2026 | Jjina-embeddings-v5-omni-nano (jinaai)as “jinaai/jina-embeddings-v5-omni-nano” | | 65.5% | |
| 201up from #21 since Oct 2, 2026 | Jjina-embeddings-v5-text-nano (jinaai)as “jinaai/jina-embeddings-v5-text-nano” | | 65.5% | |
| 221up from #23 since Oct 2, 2026 | gte-Qwen2-7B-instruct (Alibaba-NLP)as “Alibaba-NLP/gte-Qwen2-7B-instruct” | | 62.5% | |
| 231up from #24 since Oct 2, 2026 | BBidirLM-1.7B-Embedding (BidirLM)as “BidirLM/BidirLM-1.7B-Embedding” | | 63.5% | |
| 241up from #25 since Oct 2, 2026 | ITICT-TIME-and-Querit-embedding-v1 (ICT-TIME-and-Querit)as “ICT-TIME-and-Querit/ICT-TIME-and-Querit-embedding-v1” | | 63.8% | |
| 251up from #26 since Oct 2, 2026 | Imultilingual-e5-large-instruct (intfloat)as “intfloat/multilingual-e5-large-instruct” | | 63.2% | |
About this board and its history
What it measures
Text embeddings across 131 tasks and 250+ languages: retrieval, clustering, classification, similarity.
Rank at each month's end
Modelsby Glyph