Open PL LLM - PolEmo2-OUT (generative, 5-shot): leaderboard

Sentiment classification of Polish online reviews from the out-of-domain PolEmo 2.0 test set, whose review domains are absent from training. Open PL LLM Leaderboard (SpeakLeash), 5-shot, generative variant: the answer is extracted from generated text; accuracy (%), higher is better. Runs not available on the Hugging Face Hub are excluded.

Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 318 models tracked.

Top models

#ModelScoreOverall rank
1Qwen 2 72B Instruct84.21#580
2Qwen 2 72B82.59#516
3Llama 3.3 70B Instruct81.78#541
4Qwen 3 32B81.78#449
5Llama 3.1 70B Instruct81.58#574
6Qwen 2.5 72B81.58#514
7cogito-v1-preview-qwen-14B81.38
8Mistral Large 2 (Jul)81.17#497
9QwQ 32B-Preview80.97#670
10Llama 4 Scout Instruct80.57#548
11Qwen 2.5 14B Instruct80.36#655
12Llama 3.1 Nemotron 70B Instruct80.16#562
13Llama 3 70B Instruct79.96#641
14Qwen 2.5 32B Instruct79.76#511
15Mistral Large 2 (Nov) Instruct (2411)79.76#481

No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #6 GPT-6.1 Sol.

Interactive version: theaggregate.ai/benchmark?slug=open-pl-llm-polemo2-out-generative-5-shot · How It Works · Data refreshed daily, snapshot 2026-11-06.