OpenAI recently released gpt-oss, their first open-weight model since GPT-2 back in 2019.

This isn't just another model release.

While there have been open models before, this is OpenAI giving us something very similar to ChatGPT, which has been setting the standard for what an LLM is for years now, as an open source model.

For the first time in years, you can download and run cutting-edge OpenAI technology completely offline on your own hardware.

I'm running the 20b model on my 2021 MacBook Pro right now.

Let me repeat that:

I have "ChatGPT" running locally on a three-year-old laptop.

You don't need powerful servers or expensive GPUs, you can run ChatGPT locally on your laptop.

• No internet connection required
• No API calls to a server
• No rate limits
• Complete privacy

Why is this important?

One thing that limited the adoption of AI in the corporate environment, is the fear that their precious data is leaving their infrastructure and running in the cloud. This is especially important for industries where privacy is critical and data must remain private. But now it's easier than ever to run a big LLM model on-premise.

• Healthcare organizations can now run medical AI without patient data ever leaving their infrastructure
• People in remote locations can access frontier AI capabilities without connectivity
• Developers can prototype and iterate without worrying about API costs or throttling
• Companies can build a product with an AI model built in without vendor lock-in or external dependencies

Technical specs

• gpt-oss-120b → matches o4-mini performance while running on a single enterprise GPU. They used an efficient architecture that allows this huge model to run on a single 80GB GPU instead of requiring multiple GPUs.
• gpt-oss-20b → delivers o3-mini quality on consumer hardware with just 16GB of memory, like a MacBook.

Both are Apache 2.0 licensed, which means you can modify, customize, and commercialize them freely.

This represents billions of dollars of research that OpenAI is now giving away.

Sam Altman called it "the best and most usable open model in the world."

Far from perfect

They might not be the best models for everyday use: my laptop fans started spinning like crazy for 5 minutes when asked a complicated question, but they are important for what they represent.

The models are a starting point, the open source community will be able to improve them and make them better. Smaller fine-tuned models, that can run faster and more efficiently, will be created thanks to this.

The bottom line

We just entered an era where world-class AI runs in your pocket, not just in the cloud.

The democratization of AI isn't coming anymore.

It's here.

If you want to run gpt-oss on your own hardware to check it out, the easiest way is to use ollama. It's a simple tool that allows you to run models locally on your own hardware, with a UI to interact with them. You can download the gpt-ossmodel easily from ollama.