family phi · 1 variants · unknown
| benchmark | category | best score |
|---|---|---|
| ARC-Challenge | knowledge | 90.8 |
| PIQA | knowledge | 88.1 |
| OpenBookQA | knowledge | 87.8 |
| GSM8K | math | 87.3 |
| BoolQ | knowledge | 83.7 |
| CommonsenseQA | knowledge | 80.8 |
| WinoGrande | knowledge | 80.1 |
| HellaSwag | knowledge | 79.6 |
| BBH | reasoning | 77.6 |
| MMLU | knowledge | 75.5 |
| MBPP | coding | 70.3 |
| TruthfulQA | safety | 69.6 |
| TriviaQA | knowledge | 66.0 |
| HumanEval | coding | 59.1 |
| MATH | math | 48.1 |
| MedQA | domain | 46.3 |
| AGIEval | reasoning | 43.9 |
| variant | kind | quant | bits | vRAM | dl | org | year |
|---|---|---|---|---|---|---|---|
| Phi-3-small-128k-instruct | original | bf16 | 16.0 | 1,095 | microsoft | 2025 |