How to Run LLM Locally on Laptop: Ollama and LM Studio Guide (2026)

Last updated: 10 October 2026

Want a ChatGPT-style assistant that works without internet, costs nothing per message, and never sends your files to a server? You can do that today on an ordinary laptop. This guide explains how to run LLM locally on laptop using two free tools, Ollama and LM Studio. You will see exactly how much RAM and GPU memory you need, which models fit on 8 GB, 16 GB and 32 GB machines, and step-by-step setup for both Windows and Mac. We also cover common errors, speed tips, and when a RAM or SSD upgrade is actually worth the money.

Quick answer: To run an LLM locally on a laptop, install Ollama or LM Studio, then download a small open model such as Gemma 4 E2B or Qwen 3.5 4B for 8 GB RAM, or Gemma 4 12B or Qwen 3.5 9B for 16 GB. The model runs fully offline after download. More RAM or GPU memory lets you run bigger, smarter models.

What does running an LLM locally mean?

An LLM (large language model) is the "brain" behind chatbots like ChatGPT. Normally the model runs on a company's servers and you talk to it over the internet. Running it locally means the model file sits on your laptop and your own CPU, GPU and RAM do the work.

Why people do this:

• Privacy: your prompts and documents stay on your machine. Useful for client files, medical notes or company code.
• Offline use: works on a train, in a village with weak Jio or Airtel signal, or during an outage.
• No monthly bill: open models are free to download. You only pay for electricity and hardware.
• Learning: developers can build apps against a local API without paying per token.

The trade-off: local models that fit on a laptop are smaller than top cloud models like those compared in our ChatGPT vs Gemini vs Claude guide. They are good for summaries, drafting, coding help and Q&A, but they can be slower and make more mistakes on hard reasoning.

A quick word on model size and "quantization"

Models are described by parameters, like 4B (4 billion) or 12B. More parameters usually means smarter but heavier. Tools like Ollama download quantized versions, which compress the model (often to about 4 bits per weight) so it fits in less memory with a small quality loss.

A rough rule of thumb: the model needs roughly its download size in RAM or VRAM, plus a few extra GB for the conversation (context) and your operating system. A 7 GB model on a 16 GB laptop is comfortable. A 14 GB model on a 16 GB laptop is tight. Memory use also grows with longer conversations, so treat this as a starting point, not a guarantee.

Hardware requirements to run LLM locally

Here is what matters, in order of importance:

1. Memory (RAM or VRAM): decides which model sizes you can load at all.
2. GPU: decides speed. An NVIDIA GPU with its own VRAM, or an Apple Silicon Mac, is much faster than CPU-only.
3. Storage: each model is a few GB to 20+ GB. An NVMe SSD loads models much faster than a hard disk.
4. CPU: matters most when there is no usable GPU. LM Studio on Windows x64 needs a CPU with AVX2 support.

Official minimum requirements

Sources: Ollama Windows docs (opens in new tab), Ollama macOS docs, and LM Studio system requirements (opens in new tab).

Practical requirements by laptop type

On Apple Silicon Macs, the CPU and GPU share "unified memory", so a 16 GB Mac can give most of that memory to the model. On Windows laptops, a model runs fastest when it fits completely inside the GPU's VRAM; otherwise part of it spills into slower system RAM.

Which models fit on 8, 16 and 32 GB laptops

The table below uses download sizes shown on the official Ollama model library pages in October 2026. Sizes vary by quantization tag, so check the exact tag before downloading.

Notes:

• Ollama also lists tags ending in -cloud. These run on Ollama's servers, not on your laptop, so they are not offline. Avoid them if privacy is your goal.
• Tags ending in -mlx are optimised for Apple Silicon Macs.
• For coding, also look at coder-specific models such as qwen2.5-coder in smaller sizes. Our list of best free AI coding tools covers editor plugins that can connect to local models.
• Model rankings change every few months. The models above were listed in the Ollama library (opens in new tab) as of October 2026, so check it for newer releases.

Ollama vs LM Studio: which should you use?

Both are free. Ollama is free and open source. LM Studio has been free for both home and work use since July 2025, according to its official blog.

Our suggestion: start with Ollama if you want the quickest setup, or LM Studio if you prefer to see every setting on screen. You can install both; they do not conflict, though each stores its own copy of models.

How to run LLM locally on laptop with Ollama (Windows)

Step 1: Check your system

1. Press Win + R, type winver, and confirm Windows 10 22H2 or newer (or Windows 11).
2. Open Task Manager → Performance to see your RAM and GPU memory.
3. If you have an NVIDIA GPU, update the driver from NVIDIA's website or GeForce app (version 551.61 or later).
4. Make sure you have at least 15–20 GB free disk space for the app plus one or two models.

Step 2: Install Ollama

1. Go to ollama.com/download (opens in new tab) and download the Windows installer (OllamaSetup.exe).
2. Run it. Admin rights are not required; it installs in your user folder.
3. After install, Ollama runs in the background (look for the icon in the system tray).

Step 3: Download and chat with a model

You can use the Ollama desktop app: open it, pick a model from the dropdown, and start typing. It downloads the model on first use.

Or use the terminal. Open Command Prompt or PowerShell and run:

ollama run qwen3.5:4b

The first run downloads the model. When you see the >>> prompt, type your question. Type /bye to exit.

Step 4: Useful commands

ollama list              # show downloaded models
ollama ps                # show models currently loaded in memory
ollama pull gemma4:12b   # download without starting a chat
ollama rm gemma4:12b     # delete a model to free disk space

Step 5: Move models to another drive (optional)

If your C: drive is small:

1. Search Windows for "environment variables" and open Edit environment variables for your account.
2. Create a new variable named OLLAMA_MODELS and set it to a folder like D:\ollama-models.
3. Quit Ollama from the tray and start it again.

How to run LLM locally on Mac with Ollama

1. Check you are on macOS 14 Sonoma or later (Apple menu → About This Mac).
2. Download the Mac app from ollama.com/download.
3. Open the .dmg and drag Ollama into the Applications folder.
4. Launch it. On first start it may ask permission to add the ollama command to your PATH; allow it.
5. Open the Ollama app and choose a model, or open Terminal and run:

ollama run gemma4:e4b

On Apple Silicon, Ollama uses the GPU automatically. On older Intel Macs it runs on CPU only, so stick to models of 4B parameters or smaller. Models are stored in the ~/.ollama folder in your home directory.

How to use LM Studio (Windows and Mac)

1. Download LM Studio from lmstudio.ai (opens in new tab). Choose Windows (x64 or ARM) or macOS (Apple Silicon).
2. Install and open it.
3. Go to the Discover (search) tab and search for a model name such as "Gemma 4" or "Qwen 3.5".
4. LM Studio shows multiple quantized files for each model. Its documentation recommends a 4-bit option or higher if your machine can run it. Pick a file whose size fits your RAM or VRAM using the rule of thumb above.
5. Click Download.
6. Open the Chat tab, select the model at the top, and start chatting.
7. Optional: open the Developer tab to start a local server, so other apps can use the model like an API.

Tip: in the model load settings you can lower the context length if you run out of memory. A shorter context uses less RAM.

Speed tips and common errors

Make it faster

• Close Chrome tabs and heavy apps before loading a model; every GB counts on 8 GB laptops.
• Keep your laptop plugged in and set Windows power mode to Best performance. Many laptops slow down the CPU and GPU on battery.
• Choose a smaller model or a more compressed (lower-bit) quantization.
• Reduce the context length if you do not need long documents.
• Keep the laptop cool. Sustained AI work heats the CPU and GPU; a basic cooling pad and a hard, flat surface help. If your laptop is already sluggish in daily use, see our Windows 11 slow laptop fix guide first.

Common errors and fixes

Should you upgrade RAM or SSD?

For local AI, memory is the single biggest upgrade. Moving from 8 GB to 16 GB lets you jump from 4B models to 9B–12B models, which is a big quality improvement. Before you buy:

1. Check if your RAM is upgradeable. Many thin laptops and all Apple Silicon Macs have soldered RAM that cannot be upgraded. Check your laptop model's spec sheet or use a tool like CPU-Z on Windows (Memory and SPD tabs).
2. Match the type and speed. DDR4 and DDR5 laptop memory (SO-DIMM) are not interchangeable.
3. Use matched pairs if you have two slots, for dual-channel speed.

Example upgrades (pick the type your laptop supports):

• 16 GB DDR4 laptop RAM (SO-DIMM)
• 32 GB (2 × 16 GB) DDR5 laptop RAM kit.

SSD: models are large, and you will want several. A 1 TB NVMe SSD gives plenty of room and fast loading. Check whether your laptop has a spare M.2 slot or needs a replacement drive.

• 1 TB NVMe M.2 SSD.
• 1 TB NVMe M.2 SSD (alternative)
• Laptop cooling pad for long sessions.

Before you buy, check your laptop maker's spec sheet for the supported RAM type and speed, the maximum RAM, and the SSD slot type (for example M.2 NVMe). Prices and listings change often.

If you are buying a new machine, our guide to the best laptop for students under ₹50,000 explains which configurations are worth it. For local AI, prioritise 16 GB RAM (upgradeable if possible) and an NVMe SSD over a slightly faster CPU.

Going further

Once a model runs locally, you can connect it to coding editors, note apps or your own scripts through the local API. Developers can also connect local models to tools through protocols like MCP; see our explainer on what an MCP server is.

Conclusion

Now you know how to run LLM locally on laptop hardware you already own. Check your RAM, install Ollama or LM Studio, and start with a model that fits: a 4B model on 8 GB, a 9B–12B model on 16 GB, or a 26B–31B model on 32 GB. If you hit memory limits often, a RAM or SSD upgrade is the most useful spend.

Next, see which cloud assistants are worth using alongside your local model in our ChatGPT vs Gemini vs Claude comparison, or browse more guides in AI How-To & Prompts.