Which AI models can I run locally on my own computer?
Last verified 2026-10-10Paid placement never changes our picks.
Short answer
As of October 2026, start with Ollama if you want a runner with a command line and local API on Mac, Windows or Linux, LM Studio if you prefer a point-and-click app with in-app model downloads, and Jan if you want a fully open-source chat app. The main catch: the model must fit in your computer's RAM or video memory.
Top picks
Simple command-line runner with a local API on all desktops
- Free tier
- Run models locally at no cost; starter cloud credits; 1 concurrent cloud request
- Starting price
- From $20
- Platforms
- macOS, Windows, Linux, API
- Last verified
- 2026-10-10
Free, MIT-licensed for local use. The $20 Pro plan only buys cloud-model usage, not anything local.
Point-and-click app to download and chat with local models
- Free tier
- Bionic agent, local LLMs, offline voice transcription, LM Link up to 5 devices,…
- Starting price
- From $20
- Platforms
- Windows; Mac/Linux see note
- Last verified
- 2026-10-10
Homepage now says LM Studio Bionic (an agent add-on). Free covers local models; $20 Bionic+ adds cloud models.
Fully open-source offline chat app with a local API
- Free tier
- Free and open source; no paid tiers listed
- Starting price
- Free
- Platforms
- Windows, macOS, Linux, API
- Last verified
- 2026-10-10
Free, Apache 2.0, no paid plan. Can also connect to online models, so check which one you selected.
Technical users wanting the engine behind most local apps
- Free tier
- Free MIT-licensed software; no account or paid plan
- Starting price
- Free
- Platforms
- Windows, macOS, Linux (build)
- Last verified
- 2026-10-10
Free MIT-licensed C/C++ project; command-line builds, steeper for beginners. Includes an OpenAI-compatible server.
A polished desktop app for local and cloud models side by side
- Free tier
- Core Studio Desktop features; several marked Limited (personas, Knowledge Stack…
- Starting price
- See pricing page
- Platforms
- Windows, macOS, Linux
- Last verified
- 2026-10-10
Core app is free; paid plans are annual or one-time, no monthly tier. Vendor says SOC 2 / ISO 27001 are not complete yet.
A free, open-source chat app that chats with your local documents
- Free tier
- Desktop app and Python SDK are free; no paid GPT4All tier
- Starting price
- Free
- Platforms
- Windows, macOS, Linux
- Last verified
- 2026-10-10
MIT licensed and free; latest release on GitHub is from February 2025, so updates are slow.
Prices come from each tool's fact block. "See pricing page" means we have not confirmed the number on the official pricing page in the last 30 days, so we do not show one.
How to choose
- If you are choosing a model → size it to your memory first. Ollama lists download sizes such as Gemma 4 E4B at about 7 to 10 GB, gpt-oss 20B at 14 GB and gpt-oss 120B at 65 GB; Ollama says gpt-oss 20B runs with as little as 16 GB of memory.
- If you plan commercial use → check the model licence, not just the app's: gpt-oss is Apache 2.0 and Llama 4 uses Meta's community licence (per Ollama pages); Gemma 4 is reported as Apache 2.0 only in news coverage, and we found no Qwen licence on the pages read. Verify on the official model card.
- If you want everything to stay on your machine → avoid Ollama models tagged ':cloud' and the paid Ollama and LM Studio plans; those cover hosted models that run on the vendor's servers.
- If you are technical and want the engine itself → llama.cpp (MIT licence, runs on CPU or many GPUs, with an OpenAI-compatible server); most desktop apps build on it.
- If a model feels slow on your laptop → step down to a 4B to 12B model before blaming the app; speed depends mostly on model size and whether it fits in GPU memory.
How we checked
- 6 of 6 picks have a Toolsfine fact block.
- 6 of 6 picks had their pricing checked on the official site in the last 30 days.
- 1 of 1 models also suggested Ollama.Panel: Claude (2026-10-10).
- Verified on 2026-10-10.
Listed in other directories
| Tool | theresanaiforthat.com | futurepedia.io | toolify.ai |
|---|---|---|---|
| Ollama | Not checked | Listed | Not checked |
| LM Studio | Not checked | Listed | Not checked |
| Jan | Not checked | Not checked | Not checked |
| llama.cpp | Not checked | Not checked | Not checked |
"Not checked" means we could not open that directory's page; it does not mean the tool is missing there.
Where sources disagree
- LM Studio name and desktop platforms: LM Studio's homepage now leads with 'Bionic'. Its download page showed us a desktop installer for Windows only, with a headless installer for Mac and Linux, while its system-requirements docs still list macOS and Linux.
- Gemma 4 licence: News coverage reports Gemma 4 as Apache 2.0, but Ollama's model page does not state a licence. Confirm on Google's model card before commercial use.
- Memory requirements: Listed sizes are download sizes. Real memory use rises with context length and depends on quantisation, so treat them as minimums.
- Jan platforms: Jan's homepage showed only a Mac download to our fetcher; its GitHub page lists Windows, macOS and Linux builds.
Also mentioned elsewhere
Other sources name these tools for this job. They are not in our picks because we have not checked them yet.
- gpt-oss (OpenAI) · Official siteOpen-weight model, not an app: 20B and 120B sizes, Apache 2.0 per Ollama's page.
- Gemma (Google) · Official siteOpen model family with small sizes and image input; confirm the licence on Google's model card.
- Qwen (Alibaba) · Official siteOpen model family including coding variants; licence not found on the pages we read.
- Llama (Meta) · Official siteLlama 4 uses Meta's community licence and is too large for most laptops; smaller Llama 3.x models exist.
- Msty Studio · Official siteDesktop workspace for local and online models; free core plan, paid Aurum tiers.
- GPT4All · Official siteFree desktop chatbot with local document chat; 2026 project activity not verified.
- Open WebUI · Official siteWeb interface often paired with Ollama; not opened for this page.
- Hugging Face · Official siteModel hub where most downloadable open weights are published.
Know a tool we should check? Submit it →
Related answers
Changelog
- 2026-10-11 · Page updated
- 2026-10-11 · First published
- 2026-10-10 · Picks, prices and sources re-checked
Last verified 2026-10-10 · next re-check 2026-11-10. Report an error
Researched with AI assistance and checked by our editors. AI can make mistakes and vendors change plans often — confirm prices and terms on the vendor's site before you buy.
Paid placement never changes our picks. Data: JSON