ko-gemma-2-9B-it: benchmark results
rtzr's Korean instruction fine-tune of Gemma 2 9B (SFT + DPO). Provider: Return Zero. Released 2024-07-11. Access: Open.
Aggregate Rating 1571 ± 18, rank #541 of 1656 rated models, from 19 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-EQ-Bench | 59.49 | EQ-Bench score | 98.8 |
| Open Ko-LLM Leaderboard | 50.98 | Average (%) | 97.6 |
| Open Korean LLM Leaderboard | 50.98 | Average Score (%) | 97.6 |
| Open Korean LLM - KorNAT-CKA | 37.97 | Normalized accuracy (%) | 97.5 |
| Open Korean LLM - Ko-IFEval | 62.89 | Strict accuracy, prompt/instruction mean (%) | 97.4 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 68.46 | Exact match, flexible extract (%) | 97.1 |
| LogicKor - Reasoning | 9.28 | Score (0-10) | 96.8 |
| Open Korean LLM - KorNAT-Helpful | 52.5 | Normalized accuracy (%) | 93.3 |
| Open Korean LLM - KorNAT-Harmless | 69.63 | Normalized accuracy (%) | 92.7 |
| LogicKor - Multi-Turn | 8.76 | Score (0-10) | 92 |
| LogicKor | 8.67 | Score (0-10) | 90.5 |
| LogicKor - Single-Turn | 8.59 | Score (0-10) | 89.7 |
Interactive version: theaggregate.ai/model?slug=ko-gemma-2-9b-it · How It Works · Data refreshed daily, snapshot 2026-10-20.