SauerkrautLM-Qwen-32B: benchmark results
A VAGOsolutions and Hyperspace.ai German-English fine-tune of Qwen1.5-32B. Its card reports SFT for two epochs on 160K examples, then DPO for one epoch on 110K examples; these sample counts are not training-token totals. Provider: VAGOsolutions. Access: Open.
Aggregate Rating 1456 ± 39, rank #3508 of 10916 rated models, from 11 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 74.19 | Accuracy (%) (5-shot) | 95.9 |
| Open Portuguese LLM - BLUEX | 67.18 | Accuracy (%) | 89.6 |
| Open Portuguese LLM - ENEM | 75.65 | Accuracy (%) | 89.5 |
| Open Portuguese LLM - OAB Exams | 55.49 | Accuracy (%) | 88.1 |
| Open LLM Leaderboard v1 - WinoGrande | 81.37 | Accuracy (%) (5-shot) | 78.9 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 61.04 | MC2 (%) (0-shot) | 77 |
| Open LLM Leaderboard v1 - GSM8K | 46.78 | Accuracy (%) (5-shot) | 60.3 |
| Open LLM Leaderboard v1 - HellaSwag | 82.15 | Normalized accuracy (%) (10-shot) | 49.2 |
| Open LLM Leaderboard v1 - ARC Challenge | 58.79 | Normalized accuracy (%) (25-shot) | 40.9 |
| Open Portuguese LLM - ASSIN2 RTE | 47 | Macro F1 (%) | 15 |
| Open Portuguese LLM - FaQuAD NLI | 39.78 | Macro F1 (%) | 7 |
Interactive version: theaggregate.ai/model?slug=sauerkrautlm-qwen-32b · How It Works · Data refreshed daily, snapshot 2026-11-08.