MAmmoTH2-8B-Plus: benchmark results
TIGER Lab (CMU)'s MAmmoTH2-8B-Plus, a Llama 3 8B model instruction-tuned on 10 million web-harvested instruction-response pairs (WebInstruct) plus additional public instruction-tuning datasets. Provider: CMU. Released 2024-05-10. Access: Open.
Aggregate Rating 1434 ± 15, rank #1058 of 1635 rated models, from 39 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 70.74 | Accuracy (%) (5-shot) | 94.1 |
| Open Korean LLM Leaderboard | 44.19 | Average Score (%) | 90.1 |
| Open Korean LLM - Ko-IFEval | 42.87 | Strict accuracy, prompt/instruction mean (%) | 87 |
| Open Korean LLM - Ko-EQ-Bench | 34.53 | EQ-Bench score | 84.4 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 48.52 | Exact match, flexible extract (%) | 82.2 |
| Open Chinese LLM - GSM8K | 55.65 | Accuracy (%) | 76.6 |
| Open Korean LLM - KorNAT-CKA | 27.51 | Normalized accuracy (%) | 70.4 |
| Open Korean LLM - KorNAT-Helpful | 48.53 | Normalized accuracy (%) | 69.6 |
| Open Korean LLM - Ko-GPQA-Diamond | 24.75 | Normalized accuracy (%) | 69.5 |
| Open Korean LLM - KorNAT-Harmless | 63.89 | Normalized accuracy (%) | 68 |
| Open LLM Leaderboard v1 - MMLU | 64.3 | Accuracy (%) (5-shot) | 67.9 |
| Open Chinese LLM - TruthfulQA MC | 55.82 | Accuracy (%) | 65.1 |
Interactive version: theaggregate.ai/model?slug=mammoth2-8b-plus · How It Works · Data refreshed daily, snapshot 2026-11-05.