Where are LM Studio models saved? Identify them before you move them

On a standard current installation, LM Studio keeps its managed library under ~/.lmstudio/models on macOS and Linux, and %USERPROFILE%\.lmstudio\models on Windows. Confirm the active library with lms ls before moving anything: imports, links and later releases can change what a runtime actually sees.

The usual LM Studio model path

• macOS and Linux: ~/.lmstudio/models
• Windows: %USERPROFILE%\.lmstudio\models

That is the managed library, not a promise that every GGUF file you have ever opened was copied there. LM Studio can import by moving, copying or linking a file. A symbolic link can point outside the library; a hard link shares underlying storage on the same filesystem. Inspect the inventory before assuming the directory tells the whole story.

Ask LM Studio what it can see

The documented lms ls command lists downloaded models, with size, architecture and parameter information. Detailed and JSON output make it a better starting inventory than a blind filesystem search.

# Human-readable inventory
lms ls

# Structured inventory for a script or audit
lms ls --json

Import local files without an accidental move

LM Studio documents lms import for bringing a local model file into its models directory. By default, import moves the file. When you need to preserve the source, use a documented copy, hard-link or symbolic-link option; run a dry run first.

# Inspect the proposed import without changing the file
lms import /path/to/model.gguf --dry-run

# Preserve the source by copying it into the managed library
lms import /path/to/model.gguf --copy

Make a storage decision after the runtime decision

LM Studio needs its active model files where it expects them. Tensor Archive does not replace that runtime directory or make the runtime discover an archive automatically. Its role begins when you have identified a completed related family you want to retain and recover later.

For example, you may keep a completed group of related local releases, test their archive and restore, then reclaim an old duplicate working copy manually. That is different from moving a live LM Studio model because the disk is full.

A safe sequence

1. Run lms ls and identify the downloaded models and their approximate disk use.
2. Use the documented import workflow and dry run before changing the location of a local file.
3. Confirm LM Studio can still load the model before removing an earlier copy.
4. For completed related artifacts, archive, verify and restore a separate copy before manual reclamation.

For another local runtime, see where Ollama stores models. For a broader first cleanup, use the model-library retention checklist.