Phi-3 Mini punches well above its weight class for a 3.8B parameter model, showing surprising competence in reasoning and instruction-following tasks that typically require much larger models. Its 128k context window is unusually generous for something this compact, making it capable of processing long documents without the memory footprint of bigger alternatives. The trade-off is that it can still stumble on complex multi-step reasoning or nuanced tasks where larger models have a clear edge.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| IFEval | 59.8 | accuracy | 4d ago |