Phi-4

What Phi-4 is for

Phi-4 is a reasoning model: it spends output tokens working through a problem before answering, which makes it slower per useful word and better on problems that need the working shown. Microsoft released it in 2024-12 with a single 14B size under MIT, with a 16K context window. Budget for the extra tokens the thinking phase consumes. The catalogue records its strength as strong 14B on reasoning benchmarks.

Sizes and memory

The 14B needs about 9 GB of weights at Q4_K_M. That is the whole range; there is no larger or smaller variant in this record. Recommended VRAM adds roughly half again on top of the weights for the KV cache and the runtime.

Size	Weights at Q4_K_M	Recommended VRAM	Smallest card in our guide that holds it
14B	9 GB	14 GB	NVIDIA RTX 3060 12GB

Hardware that runs Phi-4

18 of the 19 cards in our hardware guide hold at least one size of Phi-4 at Q4_K_M. The cheapest that does is the NVIDIA RTX 3060 12GB at about $250, running the 14B.

The largest size any of them holds is the 14B at 9 GB, on the NVIDIA RTX 3060 12GB (12 GB at 360 GB/s). Bandwidth, not capacity, sets the speed once a size fits, so a card further down the list with faster memory will generate more tokens per second on the same weights than a larger, slower one.

What people use it for

The catalogue lists 3 recorded uses for Phi-4: chat, reasoning, math. It is tagged chat, reasoning, small.

Where Phi-4 sits in the Phi family

Our catalogue holds 4 Phi entries. Ordered by release date, Phi-4 (2024-12) is the 2nd of them, and Phi-4 Reasoning (2025-04) is newer. Phi-4 Reasoning carries the same recorded memory footprint, so choosing between it and this one is a question of behaviour rather than of hardware: Phi-4 is the one recorded here for strong 14B on reasoning benchmarks.

The MIT licence puts no usage ceiling on it, which matters if the thing you are building has users. That is the practical difference between this and a vendor community licence.

Context window: 16K

16K tokens comes to roughly 12,000 words at the usual three-quarters-of-a-word-per-token ratio: a long chat, a single source file, one chapter. It is the band where a retrieval step stops being optional, because the interesting documents do not fit. The recommended VRAM figure above assumes a working context rather than the maximum, and filling this one will push past it.

How to run Phi-4 locally

Our catalogue does not record registry tags, so look the current tag up in the Ollama library or on Hugging Face before pasting these. The commands below are the shape of the workflow, not a copy-and-paste recipe.

Option 1: Ollama (simplest)

# find the tag for Phi-4 at ollama.com/library
ollama run <tag>

Option 2: Mullama (production)

mullama pull <tag>
mullama run <tag>

Option 3: llama.cpp (CLI)

# download a GGUF from Hugging Face, searching for "Phi-4 GGUF"
./llama-cli -m model.Q4_K_M.gguf -p "Hello, AI!"

Option 4: Python with Mullama or llama-cpp-python

from mullama import Model, Context
model = Model.load("model.Q4_K_M.gguf", n_gpu_layers=99)
ctx = Context(model, n_ctx=4096)
print(ctx.generate("Hello, AI!", 256))

Sources

Model record from src/data/models.json (2026-06-29); card capacities from src/data/gpus.json (2026-06-29). Upstream: Curated from ollama.com/library + community benchmarks + paperswithcode + huggingface.