TildeOpen-30B: benchmark results
Tilde.ai's TildeOpen-30B, a 30B dense base model trained on the LUMI supercomputer with an equitable tokenizer and curriculum learning to serve 19 underrepresented Nordic and Eastern European languages. Provider: Tilde. Released 2025-08-19. Access: Open.
Aggregate Rating 1390 ± 27, rank #1239 of 1635 rated models, from 51 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open PL LLM - 8TAGS (multiple choice, 5-shot) | 76.88 | Accuracy (%) | 73.7 |
| Open PL LLM - CBD (multiple choice, 5-shot) | 24.41 | Macro F1 (%) | 55.4 |
| Open PL LLM - CBD (generative, 5-shot) | 24.29 | Macro F1 (%) | 50.3 |
| Open PL LLM - PolEmo2-IN (multiple choice, 5-shot) | 75.9 | Accuracy (%) | 46.4 |
| Polish Knowledge Benchmark - Cuisine | 78 | Accuracy (%) | 45.6 |
| Polish Knowledge Benchmark - Law & Administration | 77.58 | Accuracy (%) | 36.8 |
| Open PL LLM - PolQA Reranking (multiple choice, 5-shot) | 60.93 | Accuracy (%) | 33 |
| Open PL LLM - Belebele (multiple choice, 5-shot) | 65.33 | Accuracy (%) | 32.8 |
| Polish Knowledge Benchmark - Religion & Philosophy | 79.34 | Accuracy (%) | 31.6 |
| Open PL LLM - PolEmo2-OUT (multiple choice, 5-shot) | 63.56 | Accuracy (%) | 31.3 |
| Polish Knowledge Benchmark - Art | 68.18 | Accuracy (%) | 30.9 |
| Open PL LLM - PPC (multiple choice, 5-shot) | 56.9 | Accuracy (%) | 29.7 |
Interactive version: theaggregate.ai/model?slug=tildeopen-30b · How It Works · Data refreshed daily, snapshot 2026-10-15.