Phi-1.5: benchmark results
Microsoft's Phi-1.5, a 1.3B transformer base model trained on the same data sources as Phi-1 plus new synthetic NLP texts, excluding generic web-crawl data and without RLHF fine-tuning. Provider: Microsoft. Released 2023-09-11. Access: Open.
Aggregate Rating 1242 ± 10, rank #1549 of 1656 rated models, from 189 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LA Leaderboard - WNLI Basque | 61.97 | Accuracy (%) | 97.8 |
| MMLU-by-task - College Mathematics | 41 | Accuracy (%) | 97.2 |
| LA Leaderboard - WiC Basque | 50.29 | Accuracy (%) | 89 |
| MMLU-by-task - Abstract Algebra | 34 | Accuracy (%) | 82 |
| AI Energy Score (Text Generation) | 5 | Energy Score (1-5) | 79.5 |
| MMLU-by-task - College Computer Science | 46 | Accuracy (%) | 75.6 |
| MMLU-by-task - Machine Learning | 38.39 | Accuracy (%) | 75.5 |
| LA Leaderboard - HumorQA | 58.82 | Accuracy (%) | 70.6 |
| MMLU-by-task - Electrical Engineering | 48.28 | Accuracy (%) | 68.5 |
| MMLU-by-task - College Physics | 27.45 | Accuracy (%) | 63 |
| MMLU-by-task - Business Ethics | 52 | Accuracy (%) | 61.8 |
| MMLU-by-task - Moral Disputes | 54.91 | Accuracy (%) | 61.5 |
Interactive version: theaggregate.ai/model?slug=phi-1-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.