Phi-3 Mini 4K Instruct: benchmark results
Microsoft's Phi-3-Mini-4K-Instruct, a 3.8B model trained on synthetic and filtered web data, post-trained with SFT and DPO; the 4K-context variant of Phi-3 Mini (128K also available). Provider: Microsoft. Released 2024-04-23. Access: Open.
Aggregate Rating 1380 ± 12, rank #1297 of 1656 rated models, from 219 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HumanLikeness - Discourse-1 | 79.7 | Humanlike Score (%) | 100 |
| HumanLikeness - Syntax-1 | 89.36 | Humanlike Score (%) | 100 |
| OpenBookQA | 88 | Accuracy (%) | 96.4 |
| HumanLikeness - Discourse-2 | 76.18 | Humanlike Score (%) | 94.7 |
| HumanLikeness - Sound-2 | 67.57 | Humanlike Score (%) | 94.7 |
| SuperCLUE-DSPSafeBench - Overall | 78.62 | Score | 90.9 |
| HumanLikeness - Meaning-1 | 76.07 | Humanlike Score (%) | 89.5 |
| HumanLikeness - Overall | 64.73 | Overall Humanlike (%) | 89.5 |
| BiGGen-Bench | 3.82 | Average Score (1-5) | 84.3 |
| ARC Challenge (AI2) | 84.9 | Accuracy (%) | 82.9 |
| Big-Bench Hard | 71.7 | Average (%) | 82.3 |
| Open LLM Leaderboard - GPQA | 33.22 | Score | 81.4 |
Interactive version: theaggregate.ai/model?slug=phi-3-mini-4k-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.