Llama-3-Smaug-8B: benchmark results
Abacus.AI's Llama-3-Smaug-8B, fine-tuned from Llama 3 8B Instruct using Abacus.AI's Smaug recipe for improving real-world multi-turn conversation quality. Provider: Abacus.AI. Released 2024-05-01. Access: Open.
Aggregate Rating 1412 ± 13, rank #1180 of 1656 rated models, from 48 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 65.77 | Accuracy (%) (5-shot) | 85.3 |
| Open Korean LLM - Ko-EQ-Bench | 33.61 | EQ-Bench score | 83.5 |
| Open Korean LLM Leaderboard | 41.28 | Average Score (%) | 81.9 |
| Open Ko-LLM Leaderboard | 41.28 | Average (%) | 81.8 |
| Open Korean LLM - Ko-IFEval | 38.11 | Strict accuracy, prompt/instruction mean (%) | 79.5 |
| Open Korean LLM - KorNAT-Harmless | 64.41 | Normalized accuracy (%) | 74 |
| Open Korean LLM - KorNAT-Helpful | 48.78 | Normalized accuracy (%) | 72.3 |
| Open LLM Leaderboard v1 - GSM8K | 55.88 | Accuracy (%) (5-shot) | 67.5 |
| Open Medical LLM - MMLU College Medicine | 64.74 | Accuracy (%) | 64.1 |
| Open Medical LLM - MedMCQA | 56.08 | Accuracy (%) | 64.1 |
| Open Medical LLM - MedQA (USMLE) | 59.31 | Accuracy (%) | 62 |
| Open Medical LLM - MMLU Medical Genetics | 78 | Accuracy (%) | 58.3 |
Interactive version: theaggregate.ai/model?slug=llama-3-smaug-8b · How It Works · Data refreshed daily, snapshot 2026-10-20.