Llama 3 8B is Meta's compact but capable text model that punches above its weight for everyday language tasks. It handles instruction-following, summarization, and light reasoning with reasonable fluency, though it can struggle with complex multi-step problems where larger models have a clear edge. Its open-weight nature means it runs locally on modest hardware, making it a practical workhorse for experimentation and fine-tuning.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| MMLU-Pro | 24.6 | accuracy | 26d ago |
| BBH | 24.5 | accuracy | 26d ago |
| MATH | 4.5 | accuracy | 26d ago |
| MuSR | 6.2 | accuracy | 26d ago |
| IFEval | 14.6 | accuracy | 26d ago |
| GPQA Diamond | 7.4 | accuracy | 26d ago |