BenchMark'd
No ratings yet
Be the first to review this model
An enhanced reasoning variant of Phi-4 that outperforms o1-mini and DeepSeek-R1-Distill-70B on key reasoning benchmarks. Uses 1.5x more inference tokens for deeper chain-of-thought reasoning. Open-weight release from Microsoft Research.
Released
May 1, 2025
Parameters
14B
Context
16K
Pricing
Free
| Benchmark | Category | Score | Performance |
|---|---|---|---|
MMLU | knowledge | 80.6% | 81 |
HumanEval | coding | 78.9% | 79 |
MATH | reasoning | 85.2% | 85 |
Last updated: March 15, 2026
Benchmark scores may vary based on evaluation methodology and conditions.