Llama-3.2-Rabbit-Ko-3B-Instruct: benchmark results
CarrotAI's Korean-and-English instruction fine-tune of Meta's Llama 3.2 3B Instruct, SFT-trained on Korean instruction data. Provider: Meta. Released 2024-09-30. Access: Open.
Aggregate Rating 1405 ± 39, rank #5844 of 10916 rated models, from 15 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - IFEval | 71.99 | Score | 85.7 |
| Open Korean LLM - Ko-EQ-Bench | 31.94 | EQ-Bench score | 81.2 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 40.18 | Exact match, flexible extract (%) | 74.7 |
| Open Korean LLM Leaderboard | 39.49 | Average Score (%) | 72.5 |
| Open LLM Leaderboard - MATH Level 5 | 20.54 | Score | 71.9 |
| Open Korean LLM - Ko-IFEval | 33.46 | Strict accuracy, prompt/instruction mean (%) | 67.1 |
| Open Korean LLM - KorNAT-CKA | 22.4 | Normalized accuracy (%) | 39.5 |
| Open LLM Leaderboard - BBH | 44.27 | Score | 32.3 |
| Open LLM Leaderboard - MMLU-Pro | 28.22 | Score | 31.5 |
| Open Korean LLM - Ko-GPQA-Diamond | 21.72 | Normalized accuracy (%) | 29.4 |
| Open Korean LLM - KorNAT-Helpful | 44.22 | Normalized accuracy (%) | 27.9 |
| Open LLM Leaderboard - GPQA | 27.1 | Score | 25.8 |
Interactive version: theaggregate.ai/model?slug=llama-3-2-rabbit-ko-3b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.