Model · Artificial Analysis
Microsoft/Phi-4
Benchmark data by Artificial Analysis.
| Author / organization | Microsoft |
|---|---|
| Access | Hosted API (no weights listed here) |
| Price per million tokens | $0.125 in · $0.5 out |
| Intelligence Index | 5.9 |
| Output speed | 44 tokens/s |
| Task | text-generation |
| Created | 12 December 2024 |
| Last updated at source | — |
Reported results
| Dataset | Metric | Value |
|---|---|---|
| Artificial Analysis | AIME | 0.1433 |
| Artificial Analysis | GPQA | 0.575 |
| Artificial Analysis | Humanity's Last Exam | 0.038 |
| Artificial Analysis | Intelligence Index | 5.9 |
| Artificial Analysis | LiveCodeBench | 0.231 |
| Artificial Analysis | MATH-500 | 0.81 |
| Artificial Analysis | MMLU-Pro | 0.714 |
| Artificial Analysis | Math Index | 18 |
| Artificial Analysis | aime 25 | 0.18 |
| Artificial Analysis | ifbench | 0.2354 |
| Artificial Analysis | lcr | 0 |
| Artificial Analysis | tau2 | 0 |
| Artificial Analysis | terminalbench hard | 0.0379 |
Reported by the model's authors in the model card. Not independently verified, and setups differ between authors, so scores from different models are not always comparable. Each value links to its source.
More from Microsoft
Silytics shows public metadata from Artificial Analysis and links to the original. Information may be incomplete or out of date. Last refreshed 15 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.