Apertus-8B-Instruct-2509 — benchmark results
Swiss AI Initiative's (ETH Zurich/EPFL/CSCS) open 8B sibling of Apertus-70B, pretrained on 15T tokens of compliant open data in 1,811 languages (September 2025). Provider: Apertus. Released 2025-09-01. Access: Open.
Unified ELO 1450 ± 10, rank #996 of 1776 rated models, from 292 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LatamBoard - Spanish XNLI | 50.76 | Score (%) | 100 |
| LatamBoard - Spanish OpenBookQA | 39.4 | Score (%) | 97 |
| LatamBoard - FLORES Bidirectional | 46.76 | Score (%) | 93.9 |
| LatamBoard - Translation Score | 47.99 | Score (%) | 93.9 |
| LatamBoard - Spanish COPA | 84.2 | Score (%) | 90.9 |
| LatamBoard - Spanish Score | 63.58 | Score (%) | 90.9 |
| EuroEval Spanish NLU - Sentiment Headlines ES | 49.08 | Sentiment classification Score (%) | 86.9 |
| LatamBoard - Spanish TeleIA | 71.43 | Score (%) | 86.4 |
| EuroEval Icelandic NLU - NQII | 55.44 | Reading comprehension Score (%) | 85.4 |
| EuroEval Croatian NLU - MMS HR | 42.47 | Sentiment classification Score (%) | 83.9 |
| EuroEval Lithuanian NLU - Atsiliepimai | 46.8 | Sentiment classification Score (%) | 83.6 |
| EuroEval Faroese NLU - FoQA | 73.38 | Reading comprehension Score (%) | 82.7 |
Interactive version: theaggregate.ai/model?slug=apertus-8b-instruct-2509 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.