Mistral7B-PairRM-SPPO: benchmark results
UCLA-AGI's 7B model, fine-tuned from Mistral-7B-Instruct-v0.2 using Self-Play Preference Optimization with a PairRM reward model over UltraFeedback prompts. Provider: UCLA. Released 2024-05-04. Access: Open.
Aggregate Rating 1376 ± 19, rank #1313 of 1661 rated models, from 22 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-GPQA-Diamond | 27.78 | Normalized accuracy (%) | 95 |
| Open Korean LLM - KorNAT-Harmless | 68.16 | Normalized accuracy (%) | 91.7 |
| Open Korean LLM - KorNAT-Helpful | 49.33 | Normalized accuracy (%) | 78.5 |
| AlpacaEval 2.0 | 30.49 | LC Win Rate (%) | 68.9 |
| Open Korean LLM - Ko-IFEval | 30.85 | Strict accuracy, prompt/instruction mean (%) | 58.2 |
| Open Korean LLM - Ko-EQ-Bench | 10.7 | EQ-Bench score | 57 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 22.9 | Exact match, flexible extract (%) | 55.7 |
| Open Portuguese LLM - ASSIN2 RTE | 89.97 | Macro F1 (%) | 53.3 |
| Open Portuguese LLM - FaQuAD NLI | 64.03 | Macro F1 (%) | 48.4 |
| Open Portuguese LLM - BLUEX | 50.07 | Accuracy (%) | 46.5 |
| Open LLM Leaderboard - IFEval | 43.55 | Score | 46.1 |
| Open Korean LLM Leaderboard | 33.95 | Average Score (%) | 45.5 |
Interactive version: theaggregate.ai/model?slug=mistral7b-pairrm-sppo · How It Works · Data refreshed daily, snapshot 2026-10-23.