Gemma-2-9B-It-SPPO-Iter3: benchmark results
UCLA-AGI's third-iteration SPPO self-play preference-optimization research checkpoint of Gemma 2 9B It. Provider: UCLA. Released 2024-06-29. Access: Open.
Aggregate Rating 1538 ± 17, rank #649 of 1656 rated models, from 22 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-EQ-Bench | 62.4 | EQ-Bench score | 99.9 |
| Open Korean LLM - Ko-IFEval | 67.39 | Strict accuracy, prompt/instruction mean (%) | 98.9 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 67.1 | Exact match, flexible extract (%) | 96.6 |
| Open Ko-LLM Leaderboard | 48.97 | Average (%) | 96.2 |
| Open Korean LLM Leaderboard | 48.97 | Average Score (%) | 96.2 |
| Open Portuguese LLM - ASSIN2 RTE | 94.36 | Macro F1 (%) | 96 |
| Open Korean LLM - KorNAT-CKA | 34.99 | Normalized accuracy (%) | 95.2 |
| Open Korean LLM - KorNAT-Harmless | 66.96 | Normalized accuracy (%) | 87.3 |
| Open Korean LLM - KorNAT-Helpful | 50.43 | Normalized accuracy (%) | 85.6 |
| Open Portuguese LLM - ENEM | 73.48 | Accuracy (%) | 85 |
| Open LLM Leaderboard - GPQA | 33.89 | Score | 84.1 |
| Open Korean LLM - Ko-Winogrande | 65.27 | Accuracy (%) | 83.7 |
Interactive version: theaggregate.ai/model?slug=gemma-2-9b-it-sppo-iter3 · How It Works · Data refreshed daily, snapshot 2026-10-20.