5 Benchmarks Tracked

Polish LLM Benchmarks
Real Performance Data

Compare language models on Polish benchmarks: PLCC, CPTU, MT-Bench-PL, EQ-Bench-PL, and Open PL LLM Leaderboard. All data live from the database.

Published Jan 1, 2025 · Updated Mar 30, 2026

Explore BenchmarksBielik TrackerPLLuM Tracker

Polish SOTA

PLCC
SOTA: Gemini-3.1-Pro-Preview — 97 (average)
CPTU-Bench
SOTA: Qwen/Qwen3.5-27B thinking (API) — 4.34 (average)
Open PL LLM
SOTA: Mistral-Large-Instruct-2411 — 69.84 (average)

Polish Benchmarks

Five benchmarks covering cultural knowledge, text understanding, general evaluation, conversation quality, and emotional intelligence in Polish.

PLCC

Polish Linguistic and Cultural Competency — tests grammar, idioms, and cultural references.

165 modelsView benchmark →
SOTA: Gemini-3.1-Pro-Preview — 97 (average)

CPTU-Bench

Complex Polish Text Understanding — measures comprehension of nuanced, multi-layered Polish texts.

93 modelsView benchmark →
SOTA: Qwen/Qwen3.5-27B thinking (API) — 4.34 (average)

Open PL LLM Leaderboard

Multi-task Polish evaluation covering reasoning, knowledge, and language understanding.

660 modelsView benchmark →
SOTA: Mistral-Large-Instruct-2411 — 69.84 (average)

Polish MT-Bench

Multi-turn conversation quality — tests dialogue coherence and context retention in Polish.

50 modelsView benchmark →
SOTA: gemma-3-27b-it — 9.28 (pl-score)

Polish EQ-Bench

Emotional intelligence in Polish — evaluates understanding of emotions and social nuances.

101 modelsView benchmark →
SOTA: Mistral-Large-Instruct-2407 — 78.07 (eq-score)

Leaderboards

Top models on each Polish benchmark. Live from the database.

PLCC

Leaderboard — averageFull →
#ModelScore
1Gemini-3.1-Pro-Preview97
2Gemini-3.0-Pro-Preview95.83
3GPT-5.4-2026-03-05 (high reasoning)92.17
4Gemini-2.5-Pro-Preview-06-0592.17
5Gemini-3-Flash-Preview91.67
6GPT-5-Pro-2025-10-06 (high reasoning)91
7GPT-5.4-2026-03-05 (low reasoning)90.50
8Grok-490.50
9Gemini-2.5-Pro-Exp-03-2589.50
10GPT-5-2025-08-0789.50

CPTU-Bench

Leaderboard — averageFull →
#ModelScore
1Qwen/Qwen3.5-27B thinking (API)4.34
2gemini-2.0-flash-0014.29
3Qwen/Qwen3.5-27B non-thinking (API)4.27
4Qwen/Qwen3.5-35B-A3B thinking (API)4.22
5Qwen/Qwen3.5-35B-A3B non-thinking (API)4.18
6deepseek-ai/DeepSeek-V3.2 (API)4.14
7deepseek-ai/DeepSeek-R1 (API)4.14
8gemini-2.0-flash-lite-0014.09
9🚧DeepSeek-V3-03244.03
10deepseek-ai/DeepSeek-V3.1 (API)4.03

Open PL LLM Leaderboard

Leaderboard — averageFull →
#ModelScore
1Mistral-Large-Instruct-241169.84
2Meta-Llama-3.1-405B-Instruct-FP869.44
3Mistral-Large-Instruct-240769.11
4Qwen2.5-72B-Instruct67.92
5Qwen2.5-72B67.38
6QwQ-32B-Preview67.01
7Qwen2.5-32B66.73
8Llama-3.3-70B-Instruct66.40
9Qwen2-72B66.02
10remek/v3/rl-instruct/110k65.99

Polish MT-Bench

Leaderboard — pl-scoreFull →
#ModelScore
1gemma-3-27b-it9.28
2Mistral-Small-3.1-24B-Instruct-25039.18
3Phi-49.07
4gemma-3-12b-it8.97
5Qwen2.5-32B-Instruct8.86
6Qwen2-72B-Instruct8.78
7Mistral-Small-24B-Instruct-25018.72
8Mistral-Large-Instruct-24078.66
9Gemma-2-27b-it8.62
10aya-expanse-32b8.62

Polish EQ-Bench

Leaderboard — eq-scoreFull →
#ModelScore
1Mistral-Large-Instruct-240778.07
2Mistral-Large-Instruct-241177.29
3Meta-Llama-3.1-405B-Instruct-FP877.23
4GPT-4o-2024-08-0675.15
5gpt-4-turbo-2024-04-0974.59
6Bielik-11B-v2.6-Instruct73.70
7🚧DeepSeek-V3-032473.46
8Mistral-Small-Instruct-240972.85
9Llama-PLLuM-70B-chat72.56
10Meta-Llama-3.1-70B-Instruct72.53

Bielik Cross-Benchmark Tracker

All Bielik versions across 5 Polish benchmarks. Data live from the database.

ModelPLCCCPTU-BenchOpen PL LLM LeaderboardPolish MT-BenchPolish EQ-Bench
Bielik-0.1Polish46.67----
Bielik-1.5B-v1.0-DPO-001-L2Polish--80.48--
Bielik-1.5B-v1.0-DPO-001-L3Polish--16.38--
Bielik-1.5B-v1.0-DPO-001-L3-copyPolish--74.70--
Bielik-1.5B-v1.0-m3Polish--16.84--
Bielik-1.5B-v1.0-m3bPolish--74--
Bielik-1.5B-v1.0-m4Polish--74.50--
Bielik-1.5B-v3Polish--35.78--
Bielik-1.5B-v3.0-InstructPolish272.3841.36--
Bielik-1.5B-v3.0-Instruct-RC042025Polish--20.50--
Bielik-1.5B-v3.0-Instruct-SFT-RC042025Polish--71.11--
Bielik-11B-v2Polish--34.37--
Bielik-11B-v2.0-InstructPolish-2.2050.187.5668.24
Bielik-11B-v2.1-InstructPolish-3.9283.649.5060.07
Bielik-11B-v2.2-InstructPolish-3.7283.809.3569.05
Bielik-11B-v2.2-M-1.2Polish--92.32--
Bielik-11B-v2.3-InstructPolish-3.9786.298.9770.86
Bielik-11B-v2.3-Instruct-AWQPolish--69.58--
Bielik-11B-v2.3-Instruct-GPTQPolish--32.85--
Bielik-11B-v2.3-Instruct.IQ1_M.gguf.IQPolish--26.85--
Bielik-11B-v2.3-Instruct.IQ2_XXS.gguf.IQPolish--43.53--
Bielik-11B-v2.3-Instruct.IQ3_XXS.gguf.IQPolish--84.35--
Bielik-11B-v2.3-Instruct.Q4_K_M.ggufPolish--86.57--
Bielik-11B-v2.3-Instruct.Q4_K_M.gguf.IQPolish--91.13--
Bielik-11B-v2.3-Instruct.Q6_K.ggufPolish--86.57--
Bielik-11B-v2.3-Instruct.Q8_0.ggufPolish--51.82--
Bielik-11B-v2.4-Instruct-MSPolish--65.51--
Bielik-11B-v2.4-Instruct-SLPolish--65.87--
Bielik-11B-v2.4-Instruct-TIPolish--37.45--
Bielik-11B-v2.5-InstructPolish-2.9163.95-72.00
Bielik-11B-v2.5-Instruct-D-GRPO_H_070Polish--64.57--
Bielik-11B-v2.5-Instruct-GRPO_010Polish--67.61--
Bielik-11B-v2.5-Instruct-GRPO_020Polish--67.76--
Bielik-11B-v2.5-Instruct-GRPO_030Polish--84.35--
Bielik-11B-v2.5-Instruct-GRPO_040Polish--64.19--
Bielik-11B-v2.5-Instruct-GRPO_050Polish--61.50--
Bielik-11B-v2.5-Instruct-GRPO_060Polish--91.04--
Bielik-11B-v2.5-Instruct-GRPO_H_010Polish--35.35--
Bielik-11B-v2.5-Instruct-GRPO_H_030Polish--68.21--
Bielik-11B-v2.6-InstructPolish-3.4191.16-73.70
Bielik-11B-v3-Base-20250730Polish--77.56--
Bielik-11B-v3.0-InstructPolish783.7369.48-71.20
Bielik-11B-v3.0-Instruct-FP8-DynamicPolish--80--
Bielik-11B-v3.0-Instruct.Q4_K_M.ggufPolish--88.44--
Bielik-11B-v3.0-Instruct.Q6_K.ggufPolish--88.44--
Bielik-11B-v3.0-Instruct.Q8_0.ggufPolish--65.82--
Bielik-2.1Polish61----
Bielik-2.2Polish62----
Bielik-2.3Polish62.17----
Bielik-2.5Polish75----
Bielik-2.6Polish72----
Bielik-4.5B-v3Polish--87.08--
Bielik-4.5B-v3.0-InstructPolish352.4656.13-53.58
Bielik-4.5B-v3.0-Instruct-SFT-RC042025Polish--54.84--
Bielik-7B-Instruct-v0.1Polish-2.1630.436.1531.26
Bielik-7B-Instruct-v0.1-GPTQPolish--66.44--
Bielik-7B-v0.1Polish--20.85--
Bielik-Minitron-7B-v3.0-InstructPolish64----
Bielik-PL-11B-v3.0-InstructPolish--66.24--
Bielik-PL-Minitron-7B-v3.0-InstructPolish--81.99--
Bielik-SOLAR-LIKE-10.7B-Instruct-v0.1Polish--31.87-34.17
minitron-Bielik-7B-v3.0-Instruct-GGUF.Q4_K_M.ggufPolish--42.09--
minitron-Bielik-7B-v3.0-Instruct-GGUF.Q6_K.ggufPolish--87.33--
minitron-Bielik-7B-v3.0-Instruct-GGUF.Q8_0.ggufPolish--44--
MSH-Lite-7B-v1-Bielik-v2.3-Instruct-Llama-PrunePolish--19.53--
MSH-v1-Bielik-v2.3-Instruct-MedIT-mergePolish--50.62--
speakleash/Bielik-Minitron-7B-v3.0-InstructPolish-3.38---

About Bielik

Bielik (Polish for "White-tailed Eagle") is developed by SpeakLeash. These models are specifically optimized for Polish language tasks with a custom APT4 tokenizer and trained on 292B+ tokens of Polish text. Apache 2.0 licensed.

SpeakLeash on HuggingFace →Bielik v2 Paper →Bielik v3 Paper →

PLLuM Cross-Benchmark Tracker

All PLLuM versions across 5 Polish benchmarks. Data live from the database.

ModelPLCCCPTU-BenchOpen PL LLM LeaderboardPolish MT-BenchPolish EQ-Bench
CYFRAGOVPL/Llama-PLLuM-8B-instructPolish-3.46---
CYFRAGOVPL/PLLuM-12B-nc-chatPolish-2.62---
CYFRAGOVPL/PLLuM-12B-nc-instructPolish-3.31---
CYFRAGOVPL/pllum-12b-nc-instruct-250715Polish-3.29---
Llama-PLLuM-70B-chatPolish463.61-4.8072.56
Llama-PLLuM-70B-chat-250801Polish62----
Llama-PLLuM-70B-instructPolish-3.33--69.99
Llama-PLLuM-8B-chatPolish342.25-6.0546.20
PLLuM-12B-chatPolish333.14-9.3052.26
PLLuM-12B-instructPolish-3.09--36.21
PLLuM-12B-nc-chatPolish70--7.55-
pllum-12b-nc-chat-250715Polish-3.46--55.17
PLLuM-12B-nc-chat-250715Polish75----
PLLuM-8x7B-chatPolish443.44-7.1045.22
PLLuM-8x7B-instructPolish-3.01--39.55
PLLuM-8x7B-nc-chatPolish733.08-3.3547.29
PLLuM-8x7B-nc-instructPolish-3.22--41.75

About PLLuM

PLLuM (Polish Large Language Universal Model) is developed by OPI (National Information Processing Institute) as part of a government-backed initiative to build open Polish AI infrastructure. Models range from 8B to 70B parameters.

PLLuM Project Page →

Resources & Links

Benchmark

Open PL LLM Leaderboard

Official leaderboard on HuggingFace Spaces

Paper

Bielik 11B v2 Technical Report

Full methodology and benchmark details

Paper

Bielik v3 Small Technical Report

APT4 tokenizer & efficiency innovations

Models

SpeakLeash on HuggingFace

All Bielik model weights and documentation

Benchmark

PLCC Benchmark

Polish Linguistic and Cultural Competency

Deep Dive

CPTU-Bench

Complex Polish Text Understanding — full benchmark breakdown on CodeSOTA

Project

PLLuM Project

Polish Large Language Model by OPI

Explore More Benchmarks

See how Polish OCR models compare, or explore our broader LLM benchmark tracking.

Polish OCR BenchmarksCPTU-Bench Deep DiveGeneral LLM Benchmarks

Get notified when these results update

New models drop weekly. We track them so you don't have to.