Phi-3.5 Vision

Phi-3.5 Vision is a Phi-family model useful for small language model, edge, and low-resource workflow evaluation.

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

Small vision-language experiments

Who should use it

Small vision-language experiments
Builders who want local or self-hosted testing options.

Common workflows

Small vision-language experiments
small workflows
edge workflows
local workflows
efficient workflows

Deployment and hardware notes

~2.5 GB at Q4_K_M (4.1B parameters) — it runs on modest GPUs and is one of the few vision models comfortable inside 8 GB alongside other work.

License and usage notes

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

Strengths

Open weights where released model option for Phi workflows.
Small vision-language experiments
MIT licensed, which makes it unusually easy to ship commercially compared with most vision-language releases of its size.

Limitations

The four-crop ceiling limits how much fine detail survives on dense pages, and its text abilities sit below same-generation text-only Phi models. It uses custom modeling code, so a runtime must trust remote code or ship the implementation.
~2.5 GB at Q4_K_M (4.1B parameters) — it runs on modest GPUs and is one of the few vision models comfortable inside 8 GB alongside other work.
Context window and limits: 131,072 tokens.
Verify the exact model card, provider docs, license, and serving support before production use.

Local workflow notes

MIT licensed, which makes it unusually easy to ship commercially compared with most vision-language releases of its size.

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

Platforms: Windows, macOS, Linux

Vision spec

Memory~2.5 GB at Q4_K_M (4.1B parameters)Image inputDynamic cropping, up to 4 crops at 144 tokens each; multi-imageContext131,072 tokens

A 4.1B vision-language model built for multi-image and short-video reasoning: it crops each image adaptively rather than resizing to one square, and can compare several images in a single prompt.

Sources to verify

Exact model card

Related resources

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

Phi family hub AI tools directory Model selection guide Model source notes Models directory

Hardware~2.5 GB at Q4_K_M (4.1B parameters)RuntimeOllama or LM Studio where supported, llama.cpp, Transformers, vLLMContext131,072 tokensLast updated2026

Exact model card →