phi-4-14B: benchmark results
Microsoft's 14B Phi-4 trained heavily on synthetic data, with weights released on Hugging Face under MIT in January 2025. Provider: Microsoft. Released 2025-01-16. Access: Open.
Aggregate Rating 1516 ± 24, rank #733 of 1656 rated models, from 23 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - GPQA | 40.35 | Score | 99.3 |
| Open LLM Leaderboard - MuSR | 50.47 | Score | 98.8 |
| Open LLM Leaderboard - BBH | 66.95 | Score | 95.3 |
| Open LLM Leaderboard - MMLU-Pro | 52.78 | Score | 92 |
| TSCG 20 Tools | 84.4 | Scenario D, 20 tools, TSCG compilation: tool-selection accur | 83.3 |
| TSCG 50 Tools | 90.3 | Scenario D, 50 tools, TSCG compilation: tool-selection accur | 83.3 |
| GSMA Open-Telco - srsRAN-Bench | 83.36 | Score (%) | 83.1 |
| Open LLM Leaderboard - MATH Level 5 | 29.38 | Score | 81.4 |
| When Context Flips, Safety Breaks | 35.9 | PacifAIst BSR (self-reported) | 63.6 |
| GSMA Open-Telco - TeleTables | 29.93 | Score (%) | 50.6 |
| GSMA Open-Telco LLM Leaderboard | 50.45 | Average Score (%) | 46.1 |
| GSMA Open-Telco - TeleLogs | 19.79 | Score (%) | 43.8 |
Interactive version: theaggregate.ai/model?slug=phi-4-14b · How It Works · Data refreshed daily, snapshot 2026-10-20.