benchmarks.site

Public leaderboard · Polish language

PL-MTEB (Polish MTEB)

Tests how well "embedding" models (the part of AI search that turns text into numbers so similar meanings can be found) work on Polish text, on tasks like finding similar sentences, sorting reviews and grouping documents.

Leaderboard

Top 3 right now

  1. 1stcodefuse-ai/F2LLM-v2-14B75.13
  2. 2ndcodefuse-ai/F2LLM-v2-8B74.61
  3. 3rdQwen/Qwen3-Embedding-8B73.54

What it measures

Polish text embedding models on retrieval, classification, clustering, pair classification and semantic similarity tasks.

How to read the numbers

Mean score over all Polish tasks (0-100, from 0-1 scale) — higher is better.

Results

25 rows · higher is better (Mean (Task))CSV
1codefuse-ai/F2LLM-v2-14B75.1371.9613.99
2codefuse-ai/F2LLM-v2-8B74.6171.487.568
3Qwen/Qwen3-Embedding-8B73.5471.667.567
5Qwen/Qwen3-Embedding-4B72.3470.534.022
6codefuse-ai/F2LLM-v2-4B73.4270.284.022
7Linq-AI-Research/Linq-Embed-Mistral71.1069.287.111
8microsoft/harrier-oss-v1-0.6b70.4768.650.596
9codefuse-ai/F2LLM-v2-1.7B72.0368.961.721
10Salesforce/SFR-Embedding-Mistral69.3168.297.111
11intfloat/multilingual-e5-large-instruct69.1368.350.56
12intfloat/e5-mistral-7b-instruct69.9468.667.111
13Alibaba-NLP/gte-Qwen2-1.5B-instruct70.0067.981.543
14Salesforce/SFR-Embedding-2_R68.8867.187.111
16codefuse-ai/F2LLM-v2-0.6B69.6366.720.596
17manu/bge-m3-custom-fr64.7463.880.568
17microsoft/harrier-oss-v1-270m66.9366.040.268
21OrdalieTech/Solon-embeddings-large-0.163.4963.010.56
22codefuse-ai/F2LLM-v2-330M66.5364.390.334
23Qwen/Qwen3-Embedding-0.6B65.5064.590.596
24omarelshehy/arabic-english-sts-matryoshka65.0764.300.56
26sdadas/mmlw-e5-small63.4163.350.118
28HIT-TMG/KaLM-embedding-multilingual-mini-instruct-v164.5763.700.494
30hotchpotch/bekko-embedding-v1-a25m61.5061.660.123
32Snowflake/snowflake-arctic-embed-l-v2.061.3160.960.568
35deepvk/USER-bge-m362.4661.270.359

Note: Benchmark MTEB(pol, v1) from the MTEB leaderboard backend API; source scores are on 0-1 scale and shown here x100 to 2 decimals. Leaderboard Space last modified 2026-11-16. Models with no mean score (not run on all tasks) were skipped. Some models may have trained on part of the tasks (zero-shot share not shown).

Source: MTEB team (Hugging Face MTEB leaderboard) · mteb-leaderboard-backend.hf.space · retrieved 2026-11-17 · license: MIT (MTEB project)

Markdown · More Polish language