OpenAI releases a new GPT-OSS model: Run a free, open-source AI on your laptop

A New Era of Local AI: OpenAI's First Open-Weight Model since 2019

Illustration: Tech Bird

After a six-year hiatus, the pioneer AI company OpenAI has launched its new open-weight AI model, GPT-OSS, marking a significant shift in its strategy. The company has introduced two distinct models: gpt-oss-120b and gpt-oss-20b. What makes these models a game-changer is that you can download them for free and run powerful AI directly on your own laptop. This is a monumental release since GPT-2 in 2019 and a clear response to the evolving needs of the AI community.

GPT-OSS: Bringing Power and Flexibility to Your PC

OpenAI, the company behind the groundbreaking ChatGPT, has released its new open-weight model, GPT-OSS. This allows developers and enthusiasts to not only freely download the model but also to customize it for their specific needs, all from the comfort of their own hardware. This move directly addresses the growing demand for local, cost-effective AI solutions, a trend that is becoming increasingly popular among developers and has been adopted by competitors.

The GPT-OSS lineup includes two models:

• gpt-oss-120b: This 117-billion-parameter model is a powerhouse. It delivers performance comparable to the existing GPT-4o model on key reasoning benchmarks and can run on a single NVIDIA GPU with 80GB of VRAM.
• gpt-oss-20b: The more accessible model, with 21 billion parameters, can run on a laptop with just 16GB of VRAM. It provides performance on par with the gpt-o3-mini model, making it a viable option for a wide range of consumer-grade hardware.

Both of the new GPT-OSS models are designed for broad compatibility. They can run on a variety of platforms and services, including Azure, Hugging Face, vLLM, Ollama, llama.cpp, LM Studio, AWS, Fireworks, Together AI, Baseten, Vercel, Cloudflare, and OpenRouter. This makes them easy to integrate and use in a broader perspective, empowering developers with unprecedented flexibility. Furthermore, both models are available under the permissive Apache 2.0 license, which allows for commercial and private use with minimal restrictions.

A Strategic Shift: Why Now?

This release represents a real shift in OpenAI's strategy. Previously, the company's CEO had expressed concerns about the security of open-source models, which was the stated reason for not releasing a similar model since GPT-2.0 in 2019. This raises the question of why the company chose to introduce it now. The primary factor compelling OpenAI to launch an open-weight AI model is the increasing demand from developers who prefer models they can own and customize, as opposed to expensive proprietary models. This shift is also accelerated by a heated competitive environment, fueled by the huge success of models like Deepseek.

Safety by Design: A Core Priority

OpenAI has explicitly addressed security in its announcement blog, stating that the GPT-OSS models leverage their "state-of-the-art approaches for safety training." They further explained, "During pre-training, we filtered out certain harmful data related to Chemical, Biological, Radiological, and Nuclear (CBRN)." The company also added that they used "deliberative alignment and the instruction hierarchy to teach the model to refuse unsafe prompts and defend against prompt injections," showcasing a proactive approach to safety.

The Value of Open-Weight Models for Developers

The popularity of open-weight models stems from the public availability of their weights, allowing for easier fine-tuning and training. Unlike proprietary models, where the specific weights are kept private and cannot be easily customized, open-weight models provide full control to the user. Another major factor is the prohibitive cost of proprietary models, which can be a significant barrier for developers, especially those in smaller organizations.

As OpenAI itself noted, "These open models also lower barriers for emerging markets, resource-constrained sectors, and smaller organizations that may lack the budget or flexibility to adopt proprietary models." This new approach not only democratizes access to cutting-edge AI but also fosters innovation and competition, benefiting the entire tech ecosystem.