Polish Legal RAG - Source Classification: leaderboard
Polish Legal RAG leaderboard by gaius-lex, source classification task: identifying the source of a Polish legal text fragment. Score 0-100, higher is better; saturated, with nearly every general model between 87 and 97. The Space has been down since August 2025 and its results file is frozen.
Metric: Classification score (0-100). Source: huggingface.co. Saturation forecast: Estimated already saturated. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 96.71 | #193 |
| 2 | GPT-4.1 | 96.39 | #258 |
| 3 | Qwen 3 235B A22B 2507 Instruct | 96.21 | #315 |
| 4 | GPT-OSS-120B | 96.17 | #346 |
| 5 | Bielik-11B-v2.5-Instruct | 96.12 | #575 |
| 6 | Llama 4 Maverick Instruct | 96.08 | #467 |
| 7 | GPT-4o | 96.05 | #357 |
| 8 | DeepSeek V3 (0324) | 96.03 | #349 |
| 9 | Kimi K2 | 95.91 | #252 |
| 10 | Llama 3.3 70B Instruct | 95.89 | #541 |
| 11 | Claude Sonnet 4 | 95.88 | #207 |
| 12 | Qwen 3 32B | 95.86 | #449 |
| 13 | Llama 4 Scout Instruct | 95.86 | #547 |
| 14 | GPT-4o Mini | 95.8 | #598 |
| 15 | GPT-OSS-20B | 95.78 | #523 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=polish-legal-rag-source-classification · How It Works · Data refreshed daily, snapshot 2026-11-05.