Ollama and LM Studio tie for first at 4.7/5, and which one you want
depends on whether you like typing or clicking. Ollama is built around
the command line: ollama run gemma4 downloads a model and drops you
straight into a chat. Its hardware page covers NVIDIA, AMD and Apple
graphics, plus other cards through Vulkan. LM Studio is a desktop app
that searches Hugging Face for you, and it runs MLX models on Apple
Silicon Macs.
Jan is a tenth of a point behind at 4.6/5. Pick it if you want an open-source desktop app with GGUF and MLX models and broad GPU support. GPT4All trails at 2.9/5. It still runs models on the processor alone, but its newest changelog entry is from February 2025.
Our picks
Scores
1–5 per criterion · total is the weighted average · select a score to see its source| # | Product | Model formats and downloads ×3 | GPU support ×3 | Local API for other apps ×2 | Release recency ×2 | Total |
|---|---|---|---|---|---|---|
| 1 | Ollama | 4 | 5 | 5 | 5 | 4.7 |
| 1 | LM Studio | 5 | 4 | 5 | 5 | 4.7 |
| 3 | Jan | 5 | 5 | 3 | 5 | 4.6 |
| 4 | GPT4All | 3 | 4 | 3 | 1 | 2.9 |
How each criterion is scored
Model formats and downloads ×3
Which model file formats the app runs (GGUF, MLX, Safetensors), and whether you can find and download models inside the app or import model files you already have.
- 5 Two or more model formats, in-app Hugging Face search and download, and import of local model files
- 4 Two or more model formats (GGUF plus MLX or Safetensors), with in-app downloads
- 3 One model format (GGUF), with in-app search and download of Hugging Face models
- 2 One model format, with in-app downloads from a fixed catalog only
- 1 No in-app model downloads; model files must be added by hand
GPU support ×3
Which graphics processors (GPUs) the project documents for speeding up models: NVIDIA, AMD, Apple Silicon and Intel, plus running on the processor (CPU) alone.
- 5 NVIDIA, AMD, Apple Silicon and Intel GPUs (for example through Vulkan), and a documented processor-only mode
- 4 NVIDIA, AMD and Apple Silicon GPUs
- 3 Two GPU families, for example NVIDIA and Apple Silicon
- 2 One GPU family, for example Apple Silicon only
- 1 Processor (CPU) only
Local API for other apps ×2
Whether other programs on your computer or network can use the local models through an API that copies OpenAI's, so tools built for OpenAI work by changing one address.
- 5 As 4, plus a second API (its own REST API or an Anthropic-compatible one) and official client libraries
- 4 OpenAI-compatible chat and embeddings endpoints that other devices on your network can reach
- 3 OpenAI-compatible chat endpoint that only listens on your own computer
- 2 A local API of its own that is not OpenAI-compatible
- 1 No local API
Release recency ×2
How recently the latest version was released, from GitHub release data or the project's own changelog, counted back from 5 November 2026.
- 5 Latest release less than 1 month old
- 4 Latest release 1 to 3 months old
- 3 Latest release 3 to 6 months old
- 2 Latest release 6 to 12 months old
- 1 Latest release more than 12 months old
Ollama starts with one word. Type ollama and it asks whether you
want to run a model or hook it up to a coding tool such as Claude Code
or Codex. One more command pulls a model from its library and you’re
chatting. Other apps can join in, too: Ollama answers on an
OpenAI-compatible address on your own machine, and one environment
variable opens it to the rest of your network.
It also offers cloud-hosted models, and its FAQ is plain about the local side: “We don’t see your prompts or data when you run locally.” A local-only mode switches the cloud features off. The latest release, v0.40.15, came out on 3 November 2026. Type, chat, done.
- GPU support for NVIDIA, AMD (ROCm) and Apple (Metal), and other cards through Vulkan
- OpenAI-compatible endpoints, its own REST API, and official Python and JavaScript libraries
- Imports GGUF files and Safetensors weights
- Searching Hugging Face inside the app is not documented; models come from Ollama's library or from files you import
- Runs on the processor only on Intel Macs
Skip it if: Anyone who wants to browse Hugging Face for models from inside the app.
LM Studio is the one to open if you’d rather click than type. You search Hugging Face from its Discover tab, download a model and start chatting, and models you got elsewhere can be sideloaded. Its offline page is the clearest of the four. Chatting, chatting with documents and running the local server don’t need the internet, and “Nothing you enter into LM Studio when chatting with LLMs leaves your device.” Model search, downloads and update checks are the parts that go online.
Here’s how it makes money. The Free plan includes local model support, and LM Studio sells two paid plans for its US-hosted cloud models: Bionic+ at $20 USD a month and Pro, with higher usage limits, at $100 USD a month. Running models locally stays free. The newest changelog entry is dated 5 November 2026. Free where it counts.
- Searches and downloads Hugging Face models in the app, and runs GGUF and MLX models
- OpenAI-compatible and Anthropic-compatible endpoints, on your computer or your network
- Documents which features need the internet; chats and documents stay on your device
- On a Mac it needs Apple Silicon; Intel-based Macs are not supported
- Intel graphics cards are not mentioned on the pages we read
- Paid cloud plans sit next to the free local app
Skip it if: Owners of Intel-based Macs.
Jan is the open-source pick that still feels like an app. Install it on macOS, Windows or Linux, grab a model from its hub, which is connected to Hugging Face, or import GGUF and MLX files you already have. When a job is too big for your laptop, it also connects to OpenAI, Anthropic, Mistral, Groq and others.
The privacy defaults are written down: “Zero data collection until you say so”, with tracking preferences chosen at first launch. It misses first place by 0.1 because its local API documents a chat endpoint but not embeddings. Updates come often. Version 0.8.8 arrived on 28 October 2026.
- Runs GGUF and MLX models, with a model hub connected to Hugging Face
- GPU support for NVIDIA, AMD and Intel Arc on Windows
- Apache 2.0 license, and no usage analytics until you allow them
- Its local API documents a chat endpoint but no embeddings endpoint on the pages we read
- On macOS it needs version 13.6 or newer, and 16GB of RAM to run 7B models well
Skip it if: Developers who need an embeddings endpoint in the local API.
GPT4All’s pitch is “No API calls or GPUs required”, and the setup in its quickstart is short: install, add a model, chat. It searches Hugging Face for llama.cpp models, and LocalDocs lets a model draw on files from your computer. Sharing your chats with the GPT4All community is opt-in, through a setting called Datalake.
It ranks last on recency. The desktop app’s changelog stops at version 3.10.0, dated 24 February 2025, which scores 1/5 on release recency and pulls its total down to 2.9/5. Its API is also the most limited of the four, since it only listens on your own computer.
- Runs on the processor, Apple Silicon (Metal), and NVIDIA and AMD GPUs through Vulkan
- Searches Hugging Face for models inside the app
- LocalDocs turns your own files into sources for your chats
- Its newest changelog entry, version 3.10.0, is dated 24 February 2025
- The API server only listens on your own computer, over HTTP
- The Linux build is x86-64 only
Skip it if: Anyone who wants an app that still gets regular releases.
Questions
Do these apps work offline?
Once a model is downloaded, all four run it on your computer. LM Studio documents the split most precisely: chatting, chatting with documents and its local server work offline, while searching and downloading models need a connection. Jan advertises full offline support, and Ollama has a local-only mode that turns off its cloud features.
Do I need a graphics card?
No. GPT4All says no GPU is required, Ollama documents how to force processor-only use, and Jan has a processor engine. A GPU makes a difference, though: LM Studio recommends at least 4GB of dedicated video memory on Windows and requires a processor with AVX2.
Can other apps use the models I run locally?
Yes. All four have a local API that copies OpenAI’s, so tools built for OpenAI can point at your computer instead. Ollama, LM Studio and Jan can also open it to other devices on your network; GPT4All only listens on your own computer.
Method
Each app was scored 1–5 against a published scale on four weighted criteria: model formats and downloads (×3), GPU support (×3), local API for other apps (×2) and release recency (×2). The ranking is the weighted average. Every score cites quotes from the project’s own README, documentation, changelog or GitHub release data, captured on 5 November 2026.
We did not install or benchmark the apps, so nothing here measures speed. The scores reflect what each project documents. How our rankings work.