Edge

MITOpen weightsUpdated August 2026Practical local

Phi-4 Mini

Small open model family useful for lightweight local prototypes and edge-oriented AI experiments.

Microsoft · Phi

Editorial review

Reviewed byOpenSourcesAI EditorialLast updatedAugust 2026SourcesExact model card

Model checkpoints, context windows, provider support, local runtime compatibility, and license terms can change quickly. Verify the exact model card before production or commercial use.

Best for

Builders testing small local models on laptops, CPUs, and constrained hardware.

Who should use it

  • Builders testing small local models on laptops, CPUs, and constrained hardware.
  • Builders who want local or self-hosted testing options.

Common workflows

  • Small local models, CPU-friendly experiments, edge apps
  • edge workflows
  • small workflows
  • local workflows

Deployment and hardware notes

A strong candidate for CPU or low-VRAM testing with quantized runtimes.

License and usage notes

MIT. Open weights. Verify the exact model card and license terms for the checkpoint or hosted provider you use.

Strengths

  • Open weights model option for Phi workflows.
  • Builders testing small local models on laptops, CPUs, and constrained hardware.
  • A small-model option for local experiments on laptops, CPUs, and low-VRAM machines.
  • Tracked as Practical local in the OpenSourcesAI model directory.

Limitations

  • Small models are easier to run but need careful prompting and evaluation for complex tasks.
  • A strong candidate for CPU or low-VRAM testing with quantized runtimes.
  • Context window and limits: 131,072 tokens (RoPE-extended).
  • Verify the exact model card, provider docs, license, and serving support before production use.

Local workflow notes

A small-model option for local experiments on laptops, CPUs, and low-VRAM machines.

Local runtimes: Ollama, LM Studio, llama.cpp, Transformers

Platforms: Windows, macOS, Linux, Laptops

Will Phi-4 Mini run on your machine?

Phi-4 Mini is 3.84B parameters and needs 4.1 GB of VRAM at Q4_K_M — 2.6 GB of weights plus 1.5 GB of runtime overhead for the inference server itself.

VRAM by quantization

QuantizationWeightsNeeds (with overhead)Quality
Q4_K_M2.6 GB4.1 GBgood
Q8_04.2 GB5.7 GBhigh
FP167.5 GB9 GBreference

Fit on common hardware at Q4_K_M

HardwareMemory the model can useSystem RAMVerdict
CPU OnlyNone (CPU only)16 GBCPU offload
RTX 4060 Laptop8 GB16 GBComfortable
RTX 3060 (12GB)12 GB32 GBComfortable
RTX 4060 Ti (16GB)16 GB32 GBComfortable

Comfortable means VRAM clears the requirement by 2 GB or more. Tight means it covers the requirement with no margin. CPU offload means the model does not fit in VRAM but system RAM is at least 1.6× the weights, so it will run at reduced speed — expect roughly 1–5 tokens per second. Figures are weights plus a fixed runtime overhead and exclude KV-cache growth, which scales with context length.

Need more hardware for Phi-4 Mini? Open the PC Builder for the 7B / 8B tier →

VRAM fit by quantization level

Enter your GPU VRAM below to see which quantization of Phi-4 Mini fits and get the Ollama run command.

Sources to verify

Related resources

Continue with model source notes, local tools, and implementation guides related to this model.

Hardware4GBRuntimeOllama, LM Studio, llama.cpp, TransformersContext131,072 tokens (RoPE-extended)Last updated2026
Exact model card →

Model ecosystem connections

Use these next-step links to move from this profile into related tools, comparisons, guides, stacks, and curated shortlists.