Phi-1.5 punches well above its weight class for a 1.4B parameter model, showing surprisingly capable reasoning and coding ability that rivals much larger models on structured tasks. It was trained on carefully curated 'textbook quality' data rather than raw web crawls, which gives it a clean, methodical reasoning style. However, its small size means it struggles with broad world knowledge and can feel narrow outside its training distribution.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| IFEval | 20.3 | accuracy | 4d ago |