LES Leaderboard – Slovene Embedding Benchmark

Lestvica embeddingov za slovenščino

Comparing embedding models on Slovene NLP tasks

37 models compared on 26 shared tasks: 8 classification, 8 clustering, and 10 retrieval tasks. Overall scores equally weight the three task families. Recent evaluations also include text simplification; that extra task is excluded here so all models are ranked on the same task set.

Rank	Model	Parameters	Context Window	Score	Classification	Clustering	Retrieval
🥇	google/gemini-embedding-2	-	8,192	62.00	62.38	29.82	93.79
🥈	google/gemini-embedding-001	-	2,048	58.12	58.97	22.61	92.77
🥉	Voyage/voyage-4-large	-	32,000	57.14	53.58	24.77	93.06
4	Voyage/voyage-multimodal-3.5	-	32,000	56.56	52.36	24.82	92.48
5	Qwen/Qwen3-Embedding-8B🤗	8B	32,768	55.72	53.33	23.53	90.30
6	Cohere/embed-v5.0-pro	-	128,000	55.66	52.20	23.50	91.28
7	Snowflake/snowflake-arctic-embed-l-v2.0🤗	0.6B	8,192	55.63	52.35	23.94	90.59
8	openai/text-embedding-3-large	-	8,192	55.58	55.71	24.37	86.66
9	codefuse-ai/F2LLM-v2-4B🤗	4B	40,960	55.42	53.91	24.68	87.66
10	BAAI/bge-m3🤗	560M	8,192	55.08	52.45	22.65	90.15
11	Qwen/Qwen3-Embedding-4B🤗	4B	32,768	55.05	51.96	24.48	88.73
12	microsoft/harrier-oss-v1-0.6b🤗	0.6B	32,768	54.69	52.11	22.91	89.06
13	perplexity-ai/pplx-embed-v1-4b🤗	4B	32,768	54.69	50.65	21.48	91.94
14	perplexity-ai/pplx-embed-v1-0.6b🤗	0.6B	32,768	54.50	51.25	22.85	89.40
15	google/embeddinggemma-2🤗	740M (270M text)	8,192	54.20	51.23	24.62	86.76
16	intfloat/multilingual-e5-large-instruct🤗	0.6B	512	54.08	52.76	23.93	85.55
17	Cohere/embed-v4.0	-	128,000	53.60	48.98	20.89	90.93
18	Alibaba-NLP/gte-multilingual-base🤗	0.3B	8192	53.52	50.64	24.27	85.63
19	openai/text-embedding-3-small	-	8,192	53.37	52.07	23.80	84.24
20	google/text-multilingual-embedding-002	–	–	52.49	50.80	18.67	88.01
21	codefuse-ai/F2LLM-v2-0.6B🤗	0.6B	40,960	52.45	50.84	23.77	82.74
22	google/embeddinggemma-300m🤗	0.3B	2,048	51.56	49.34	23.04	82.31
23	jinaai/jina-embeddings-v5-text-nano🤗	0.24B	8,192	50.85	38.63	27.04	86.89
24	intfloat/multilingual-e5-small🤗	0.1B	512	50.41	47.64	20.85	82.72
25	Qwen/Qwen3-Embedding-0.6B🤗	0.6B	32,768	50.11	46.61	21.16	82.55
26	sentence-transformers/paraphrase-multilingual-mpnet-base-v2🤗	0.3B	512	46.27	51.15	21.28	66.40
27	sentence-transformers/LaBSE🤗	0.5B	512	44.71	48.85	19.86	65.43
28	cjvt/GaMS-1B🤗	1B	2,048	43.30	55.29	27.83	46.78
29	google/text-embedding-005	–	–	41.91	39.52	15.41	70.80
30	nvidia/Nemotron-3-Embed-1B-BF16🤗	1B	32,768	41.68	41.80	17.54	65.69
31	intfloat/e5-large-v2🤗	0.3B	512	39.79	39.85	15.49	64.03
32	Snowflake/snowflake-arctic-embed-l🤗	0.3B	8,192	37.55	38.25	13.62	60.77
33	NovaSearch/stella-en-400M-v5🤗	0.4B	8,192	35.43	41.60	17.92	46.78
34	intfloat/e5-small-v2🤗	~33M	512	33.93	35.95	14.15	51.70
35	cjvt/GaMS-2B🤗	2B	-	29.87	50.13	21.39	18.08
36	EMBEDDIA/sloberta🤗	~110M	512	28.29	49.81	18.55	16.52
37	sentence-transformers/all-MiniLM-L6-v2🤗	22.7M	512	25.67	32.66	14.60	29.76

🥇 1st🥈 2nd🥉 3rdBold = best in column

Scores are shown as percentages

About the Overall score

The overall score is the mean of the main Classification, Retrieval, and Clustering scores.

📝Read the Blog Post

Lestvica embeddingov za slovenščino

Thanks to Our Data Providers