Llama 3 8B Synthia V3.5: benchmark results
migtissera's Llama-3-8B-Synthia-v3.5, a Llama 3 8B fine-tune on the Synthia-v3.5 dataset of varied system contexts, tuned for instruction following and long-form conversation; uncensored. Provider: migtissera. Released 2024-05-17. Access: Open.
Aggregate Rating 1428 ± 15, rank #1108 of 1656 rated models, from 30 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-GPQA-Diamond | 27.27 | Normalized accuracy (%) | 92.4 |
| Open Ko-LLM Leaderboard | 41.99 | Average (%) | 84.3 |
| Open Korean LLM Leaderboard | 41.99 | Average Score (%) | 84.3 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 46.63 | Exact match, flexible extract (%) | 80.9 |
| Open LLM Leaderboard v1 - MMLU | 65.02 | Accuracy (%) (5-shot) | 80.5 |
| Open Korean LLM - Ko-IFEval | 36.79 | Strict accuracy, prompt/instruction mean (%) | 76.6 |
| Open LLM Leaderboard v1 - GSM8K | 63 | Accuracy (%) (5-shot) | 76.3 |
| Open Korean LLM - Ko-EQ-Bench | 18.84 | EQ-Bench score | 65.1 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 54.6 | MC2 (%) (0-shot) | 63 |
| Open Chinese LLM - TruthfulQA MC | 55.3 | Accuracy (%) | 62.4 |
| Open Korean LLM - Ko-Winogrande | 61.48 | Accuracy (%) | 62.4 |
| Open Korean LLM - KorNAT-Helpful | 47.85 | Normalized accuracy (%) | 61.8 |
Interactive version: theaggregate.ai/model?slug=llama-3-8b-synthia-v3-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.