OpenAI has announced new models that are Open Source and available for anyone to download and install on their own computers: GPT-Oss . With these models now released, it’s the perfect time to implement AI directly within your premises, in other words, executing it on your computer . Today, we will guide you through the steps to install and use these models effectively.
Differences Between the Two Models
Even though they share a similar name, GPT-Oss-120B and GPT-Oss-20B are distinctly different and come with varying requirements. The primary model, GPT-Oss-120B , is comparable to OpenAI’s O4-mini, demanding at least 60 GB of graphics memory .
Having your own ChatGPT at home is easy, but it requires a capable team | Image: Xataka
The smaller counterpart, GPT-Oss-20B , is less powerful, akin to O3-mini according to OpenAI, and can be run on edge devices . This means it can operate on your personal computer, provided it meets the minimum requirement of 16 GB of memory , preferably graphics memory.
In summary:
- GPT-Oss-120B: A large model needing at least 60 GB of VRAM or unified memory and is not suited for consumer computers.
- GPT-Oss-20B: A smaller model requiring 16 GB of VRAM or unified memory, ideal for consumer computers.
For this guide, we will focus on using GPT-Oss-20B .
Considerations to Take Into Account
Running AI locally is an intensive process , and may cause your computer to slow down significantly . While it is technically possible to execute the model with only 16 GB of RAM, it is recommended to have a high-end GPU for optimal performance.
If your computer has less than 16 GB of VRAM, the tool will resort to using RAM, which should also be equal to or greater than 16 GB. Otherwise, the system may not function correctly. As a general piece of advice, it’s ideal to allocate as many resources as possible to the model’s execution, so it is wise to close any unnecessary applications.
Installing Ollama on Your Computer
OLLAMA Installation | Image: Xataka
For this tutorial, we will use a well-established application: Ollama . This Open Source platform significantly simplifies the installation, access, and use of LLMs (Large Language Models) . Think of it as a model executor.
ChatGPT operates online and allows users to interact with a model like GPT-4O, while Ollama provides the same capabilities, but at home using models installed on your computer . It is free, open-source software available for Windows, Mac, and Linux.
Download GPT-Oss
Once Ollama is downloaded and installed on your computer, you’ll encounter an interface resembling this. Alternatively, you can also opt for the terminal interface, but the graphical interface is significantly more user-friendly.
Main interface of Ollama | Image: Xataka
In the lower right corner, you’ll find a dropdown with the name of the model you’re currently using or will be using.
Access to the different AI models from Ollama | Image: Xataka
By clicking on the dropdown, you’ll access several models, including Deepseek R1, Gemma, or Qwen. Here, we want to select “GPT-Oss: 20B“.
Download of the model, arm yourself with patience | Image: Xataka
After selecting “GPT-Oss: 20B,” just send a message in the chat to initiate the model’s download. Be patient because the model size is 12.8 GB , and downloading may take some time.
Talking to GPT-Oss-20B through Ollama | Image: Xataka
Once installed, you can start conversing with the AI just as you would with ChatGPT. However, if your GPU doesn’t meet the minimum requirements, you may notice that it operates much slower than ChatGPT. This difference is due to the fact that you’re executing the model on your own machine rather than accessing a powerful data center filled with the latest Nvidia dedicated GPUs.
Another Option: LM Studio
LM Studio | Image: Xataka
While Ollama is intuitive and straightforward, another comprehensive program you might consider is [LM Studio](https://lmstudio.ai/download) . This application is accessible for Windows, Linux, and Mac, allowing it to manage multiple models, including GPT-Oss: 20B .
LM Studio provides more advanced options for fine-tuning both your computer’s settings and the model’s behavior, but fully leveraging it may require more advanced knowledge.
In conclusion, with the introduction of GPT-Oss models from OpenAI, you can now harness the power of advanced AI directly on your computer. Whether you choose Ollama for its simplicity or LM Studio for its expansive capabilities, both offer exciting opportunities to engage with AI in new and personalized ways.
Cover image | Xataka
In Xataka | How to move from image to video using artificial intelligence: 14 essential free tools