Apertus-70B-Instruct-2509: benchmark results
Switzerland's fully open 70B instruct model from ETH Zurich, EPFL and CSCS (September 2025), with transparent training on 15T tokens across 1,800+ languages. Provider: Apertus. Released 2025-09-02. Access: Open.
Aggregate Rating 1487 ± 16, rank #820 of 1656 rated models, from 452 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Dutch Summarization - Wiki Lingua NL | 36.71 | chrF3++ (%) | 100 |
| EuroEval Icelandic Summarization - RRN | 40.02 | chrF3++ (%) | 100 |
| EuroEval Portuguese Summarization - Publico | 34.74 | chrF3++ (%) | 100 |
| EuroEval Catalan Summarization - Dacsa CA | 39.83 | chrF3++ (%) | 98.7 |
| EuroEval Czech Summarization - Czech News | 29.9 | chrF3++ (%) | 98.7 |
| EuroEval Hungarian Summarization - Hunsum | 33.96 | chrF3++ (%) | 98.6 |
| EuroEval Lithuanian Summarization - Lrytas | 34.41 | chrF3++ (%) | 98.6 |
| EuroEval English Summarization - CNN Dailymail | 44.93 | chrF3++ (%) | 97.3 |
| EuroEval Romanian Summarization - Sumo RO | 37.51 | chrF3++ (%) | 97.3 |
| EuroEval Estonian Summarization - ERR News | 31.94 | chrF3++ (%) | 96.6 |
| SEA-HELM (Malay) - Toxicity Detection | 16.57 | Normalized Score | 96.5 |
| EuroEval Swedish Summarization - Swedn | 38.69 | chrF3++ (%) | 96.1 |
Interactive version: theaggregate.ai/model?slug=apertus-70b-instruct-2509 · How It Works · Data refreshed daily, snapshot 2026-10-20.