Model
Model explorer

Phi-4-reasoning (14B)

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

14B reasoning specialist SFT-tuned on 1.4M STEM prompts with o3-mini-distilled chain-of-thought; beats DeepSeek-R1-Distill-70B on AIME despite far fewer params; shipped alongside Phi-4-reasoning-plus.

ReasoningCodingVisionFunction callingTool useAgentic
1292.1
Elo · rank #158
Parameters
14B
Active params
14B (dense)
Context
32K tokens
Architecture
Dense decoder-only Transformer (14B)
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%#92
best: GPT-5.2 · 100.0%
Arena-HardHuman preference73.3%#32
best: Qwen3-235B-A22B (Non-Thinking) · 96.1%
CodeforcesCoding1736#30
best: DeepSeek-V4-Pro (Think Max) · 3206
GPQA DiamondReasoning65.8%#141
best: GPT-5.6 · 94.6%
HumanEval+Coding92.9%#3
best: Mistral Small 3.2 (24B) · 92.9%
IFEvalReasoning83.4%#77
best: Gemma 4 26B A4B · 98.5%
LiveCodeBenchCoding53.8%#114
best: DeepSeek-V4-Pro (Think Max) · 93.5%
MMLU-ProKnowledge74.3%#79
best: Claude Fable 5 · 91.5%
OmniMathMath76.6%#2
best: Phi-4-reasoning · 76.6%
Run it locally
VRAM @ Q4
10 GB
VRAM @ FP16
29 GB
Fits on (Q4)
RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
No independently verified RTX 4090 llama.cpp benchmark found; third-party estimates cluster around 70-90 tok/s at Q4/Q5 for this 14B dense model.
Quantizations
GGUF Q4 · MLX 4-bit
Phi-4 family
Elo progression across releases
Phi-4 (14B)Dec 20241016.7Phi-4-mini (3.8B)Feb 2025529.8Phi-4-mini-instructFeb 2025523.2Phi-4-multimodal (5.6B)Feb 2025842.4Phi-4-multimodal-instructFeb 2025842.4Phi-4-reasoningApr 20251292.1Phi-4-reasoning (14B)Apr 20251292.1
Resources
HFmicrosoft/Phi-4-reasoningDOCSazure.microsoft.com/en-us/products/phiAPIAvailable via Microsoft API
API price weights · each benchmark row carries its own source badge (see methodology)