Model

Phi-4-reasoning

OPEN

Microsoft · Phi-4 family · released Apr 30, 2025

14B dense reasoning specialist SFT-tuned on ~1.4M o3-mini CoT traces; beats DeepSeek-R1-Distill-70B on math/science benchmarks. Shipped alongside Phi-4-reasoning-plus (same size, extra RL, higher scores).

ReasoningCodingVisionFunction callingTool useAgentic

1292.1

Elo · rank #157

Parameters

14B

Active params

14B (dense)

Context

32K tokens

Architecture

Dense decoder-only Transformer, 14B (same base architecture as Phi-4)

License

MIT

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

AIMEMath75.3%#91

best: GPT-5.2 · 100.0%

Arena-HardHuman preference73.3%#31

best: Qwen3-235B-A22B (Non-Thinking) · 96.1%

CodeforcesCoding1736#29

best: DeepSeek-V4-Pro (Think Max) · 3206

GPQA DiamondReasoning65.8%#140

best: GPT-5.6 · 94.6%

HumanEval+Coding92.9%#2

best: Mistral Small 3.2 (24B) · 92.9%

IFEvalReasoning83.4%#76

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding53.8%#113

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MMLU-ProKnowledge74.3%#78

best: Claude Fable 5 · 91.5%

OmniMathMath76.6%#1

best: this model · 76.6%

Run it locally

VRAM @ Q4

13.8 GB

VRAM @ FP16

31 GB

Fits on (Q4)

RTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Throughput data unavailable.

Quantizations

GGUF · MLX

Phi-4 family

Elo progression across releases

Phi-4 (14B)Dec 20241016.7 Phi-4-mini (3.8B)Feb 2025529.8 Phi-4-mini-instructFeb 2025523.2 Phi-4-multimodal (5.6B)Feb 2025842.4 Phi-4-multimodal-instructFeb 2025842.4 Phi-4-reasoningApr 20251292.1 Phi-4-reasoning (14B)Apr 20251292.1

Resources

HFmicrosoft/Phi-4-reasoning DOCSarxiv.org/abs/2504.21318 APIAvailable via Microsoft API

API price weights · each benchmark row carries its own source badge (see methodology)