gemma2-9B-cpt-sahabatai-v1-instruct: benchmark results
GoTo/AI Singapore's Sahabat-AI Gemma2 9B Instruct, an Indonesian-focused chat model continually pretrained from Gemma 2 9B and instruction-tuned on Indonesian, Javanese, Sundanese and English data. Provider: Sahabat-AI. Released 2024-11-06. Access: Open.
Aggregate Rating 1552 ± 20, rank #608 of 1656 rated models, from 23 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Sahabat-AI - Indonesian (ID) | 63.96 | Score (%) | 100 |
| Sahabat-AI - Indonesian Average | 59.91 | Average Score (%) | 100 |
| Sahabat-AI - Javanese (JV) | 60.1 | Score (%) | 100 |
| Sahabat-AI - Sundanese (SU) | 55.66 | Score (%) | 100 |
| Open LLM Leaderboard - MuSR | 47.79 | Score | 93.9 |
| Open LLM Leaderboard - GPQA | 33.47 | Score | 82.4 |
| Open LLM Leaderboard - BBH | 59.55 | Score | 81.2 |
| Open LLM Leaderboard - IFEval | 65.51 | Score | 77.8 |
| Open LLM Leaderboard - MMLU-Pro | 42.64 | Score | 77.4 |
| CultureTalk-ID - Translation into Local Languages | 21.7 | BLEU-4 (0-100) of zero-shot dialogue translation from Indone | 75 |
| Open LLM Leaderboard - MATH Level 5 | 20.54 | Score | 72 |
| CultureTalk-ID - Translation from Local Languages | 39.2 | BLEU-4 (0-100) of zero-shot dialogue translation from the lo | 62.5 |
Interactive version: theaggregate.ai/model?slug=gemma2-9b-cpt-sahabatai-v1-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.