# Best AI model for embeddings — October 2026

As of Oct 14, 2026, Microsoft Harrier Oss V1 27B & Tencent Kalm Embedding Gemma3 12B 2511 & Qwen Qwen3 Embedding 8B are tied for state of the art in Embeddings (since Apr 1, 2026; sources MTEB Multilingual v2 (MMTEB); methodology v1.0).

Category board: https://isitsota.com/c/embeddings

## Top 10

- #1 Microsoft Harrier Oss V1 27B
- #1 Tencent Kalm Embedding Gemma3 12B 2511
- #1 Qwen Qwen3 Embedding 8B
- #2 Bytedance Seed1 6 Embedding 1215
- #3 Nvidia Llama Embed Nemotron 8B
- #4 Qwen Qwen3 Embedding 4B
- #5 Microsoft Harrier Oss V1.0 6B
- #6 Codefuse AI F2llm V2 14B
- #7 Google Gemini Embedding 001
- #8 Codefuse AI F2llm V2 8B

## Crown changes

- Apr 1, 2026 — Embeddings: Qwen Qwen3 Embedding 8B and Tencent Kalm Embedding Gemma3 12B 2511 and Microsoft Harrier Oss V1 27B are co-SOTA
- Dec 23, 2025 — Embeddings: Qwen Qwen3 Embedding 8B and Qwen Qwen3 Embedding 4B and Google Gemini Embedding 001 and Nvidia Llama Embed Nemotron 8B and Tencent Kalm Embedding Gemma3 12B 2511 and Bytedance Seed1 6 Embedding 1215 are co-SOTA
- Sep 4, 2025 — Embeddings: Qwen Qwen3 Embedding 8B and Qwen Qwen3 Embedding 4B and Google Gemini Embedding 001 are co-SOTA

## Weights

- MTEB Multilingual v2 (MMTEB): primary, weight 1
- MTEB English v2: secondary, weight 0.45