Leaderboard / Microsoft (Phi)

Phi-4

Name: Phi-4
Price: 0.07 USD
Author: Microsoft (Phi)

Microsoft (Phi) · Phi-4 · released 2024-12-12 · data checked 2026-11-01 Needs verification

Phi-4 is a 14.7B-parameter dense model from Microsoft (Phi) (United States), released 2024-12-12 under the MIT license. It scores 44.7 on the Small LM Index, #16 of 26 small models. It needs about 11.4 GB of memory at 4-bit with an 8K-token context and costs $0.070 per million tokens through the cheapest API provider.

Small LM Index

44.7

#16 of 26 scored small models

Likely 34.5–54.9 · from 2 of 6 core benchmarks

Parameters

14.7B

Dense

Price per 1M tokens

$0.070

$0.056 in / $0.11 out · blended 3:1

Context window

16K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best small model on that benchmark. Sources: 2 reported by Microsoft (Phi). How we label sources.

MMLU-Pro Lab-reported Knowledge70.4

Class best: Qwen3.5 9B 82.5

GPQA Diamond Lab-reported Knowledge56.1

Class best: Qwen3.5 9B 81.7

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

Humanity's Last Exam≈ 5

AIME≈ 61

LiveCodeBench≈ 50

SWE-bench Verified≈ 25

Specifications

Lab: Microsoft (Phi) (US · United States)
Architecture: Dense transformer
Total parameters: 14.7B
Active parameters: All (dense)
Size class: Small
Context window: 16K tokens
Input modalities: text
Reasoning: Instruct (no thinking)
License: MIT
Openness: Open weights
Memory to run: ~11.4 GB at 4-bit, 8K context (more sizes)
Hugging Face: 327K downloads (30 days) · 2.3K likes
API providers: 1 on OpenRouter
API price: $0.056 in / $0.11 out per 1M (OpenRouter, 2026-11-03)

Weights on Hugging Face ↗Try on OpenRouter ↗

Compare with

Phi-4 vs Apriel 1.6 15B Thinker (14.9B, index 60.9)
Phi-4 vs Ministral 3 14B (14B, index 60.0)
Phi-4 vs Gemma 4 12B (11.9B, index 56.9)
Phi-4 vs Phi-4-reasoning-plus (14.7B, index 54.4)
Phi-4 vs LLaDA2.2 mini (16.3B, index 29.7)
Phi-4 vs Gemma 3 12B (12.2B, index 28.4)

Sources

https://huggingface.co/microsoft/phi-4

Something wrong here? Report it.

Run it locally

Uses unsloth/phi-4-GGUF (Q4_K_M). Swap the tag for another size: F16, Q2_K, Q2_K_L, Q3_K_M, Q4_K_M, Q5_K_M, Q6_K, Q8_0.

Ollama

ollama run hf.co/unsloth/phi-4-GGUF:Q4_K_M

llama.cpp

llama-server -hf unsloth/phi-4-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model mlx-community/phi-4-4bit

vLLM (GPU server)

vllm serve microsoft/phi-4

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	16K context
16-bit F16	31 GB	33 GB
8-bit Q8_0	17 GB	20 GB
4-bit Q4_K_M	11 GB	13 GB

Check it against your GPU, Mac or phone →