Which LM Studio model should I try first?

Phi-4-reasoning is the cleanest first stop here because it has a tracked LM Studio path and a practical hardware target.

When should I choose LM Studio?

People who want a desktop app, a model catalog, and a lower-friction testing loop.

Should I benchmark before using LM Studio?

Yes. The benchmark tells you which size band is realistic for your exact machine, then this page tells you which model is the cleanest first pull inside that runtime.

best model

Best local models for LM Studio

Best for people who want a graphical model browser and easy GGUF pulls. This page ranks cleaner starting points first, then links you into the model pages when you need exact memory and hardware guidance.

starter pickPhi-4-reasoning

tracked models11

tier span3B to Frontier MoE

runtime tradeoffSome advanced tuning paths still push you back toward llama.cpp or runtime-specific docs.

benchmark first

Benchmark before you commit to a LM Studio download.

Some advanced tuning paths still push you back toward llama.cpp or runtime-specific docs. The benchmark is still the fastest way to confirm whether this machine belongs in the size band that makes LM Studio feel worth using.

Benchmark this device for LM Studio Browse model pages

start here

Best first models for LM Studio

Phi-4-reasoning

13B class • 8.5 GB minimum

LM Studio has a dedicated model page and compatible GGUF packaging.

Open model page Open source

gpt-oss-20b

34B class • 15.5 GB minimum

LM Studio lists the model directly and surfaces compatible GGUF packaging.

Open model page Open source

Gemma 3 12B

13B class • 11.0 GB minimum

LM Studio lists compatible GGUF and MLX sources for this model on its catalog page.

Open model page Open source

why this runtime

What you are choosing with LM Studio

Best for: People who want a desktop app, a model catalog, and a lower-friction testing loop.
Tradeoff: Some advanced tuning paths still push you back toward llama.cpp or runtime-specific docs.
Benchmark flow: Use the benchmark first when the question is about your machine, then use this page to choose the cleanest first pull inside LM Studio.

broader catalog

More tracked models for LM Studio

Granite 4.0 Micro

3B class • 2.5 GB minimum

IBM's smallest Granite 4.0 instruct release is a pragmatic US-origin starter for local chat, extraction, and agent scaffolding.

Open model page

OLMo 3 Instruct 7B

7B class • 5.0 GB minimum

Ai2's 7B instruct release is the clearest Apache-licensed American alternative to Llama when you want a smaller fully open local model.

Open model page

Llama 3.1 8B

7B class • 6.5 GB minimum

Meta's 8B instruct release remains the safest broad-compatibility US local model when you want maximum runtime coverage.

Open model page

OLMo 3.1 Instruct 32B

34B class • 19.5 GB minimum

OLMo 3.1 32B is the strongest Apache-licensed American 32B-class option, but it asks for more memory than gpt-oss-20b to reach a clean first run.

Open model page

Granite 4.0 H-Small

34B class • 19.5 GB minimum

Granite 4.0 H-Small is a credible American midrange choice for RAG-heavy work, but it is more specialized than the general-purpose winners above it.

Open model page

Llama 3.3 70B

70B class • 40.0 GB minimum

Llama 3.3 70B remains the most established American large-model jump for a serious local machine that can already hold a 70B-class package.

Open model page

Model pages to open next

P1Static

Model page

Can I run Phi-4-reasoning locally?

Search intent: phi-4-reasoning can i run it

13B class start • 8.5 GB minimum

Microsoft

Open page

P1Static

Model page

Can I run gpt-oss-20b locally?

Search intent: gpt-oss-20b can i run it

34B class start • 15.5 GB minimum

OpenAI

Open page

P1Static

Model page

Can I run Gemma 3 12B locally?

Search intent: gemma 3 12b can i run it

13B class start • 11.0 GB minimum

Google

Open page

runtime smoke

Monthly runtime smoke matrix

Each row installs or updates the tracked runtime, downloads the starter model, and proves one local inference with the pinned prompt bundle.

These rows use hosted CPU runners so stale guidance is visible before the public install copy drifts too far from reality.

LM Studio

Runtime guidance currently needs review

Last verified: Not yet verified

Tested runtime version: Not yet verified

Monthly smoke cadence (31-day review window)

Prompt bundle: 2026.03-reference-lm-prompts-v1

Linux

GitHub-hosted Ubuntu x64 CPU runner

Install recipe: Install the latest LM Studio headless runtime (`llmster`) with the official Unix installer.

Last verified: Not yet verified

Tested version: Not yet verified

Model pull: Granite 4.0 Micro

Stale: No successful monthly smoke run recorded yet.

Open install source Open model source

macOS

GitHub-hosted macOS CPU runner

Install recipe: Install the latest LM Studio headless runtime (`llmster`) with the official Unix installer.

Last verified: Not yet verified

Tested version: Not yet verified

Model pull: Granite 4.0 Micro

Stale: No successful monthly smoke run recorded yet.

Open install source Open model source

Windows

GitHub-hosted Windows x64 CPU runner

Install recipe: Install the latest LM Studio headless runtime (`llmster`) with the official PowerShell installer.

Last verified: Not yet verified

Tested version: Not yet verified

Model pull: Granite 4.0 Micro

Stale: No successful monthly smoke run recorded yet.

Open install source Open model source