Introducing GPT-OSS: OpenAI's massive open-weight, open-source AI models featuring 20 billion and a staggering 120 billion parameters
OpenAI has made a significant move in the world of AI by releasing its first open-weight large language models in over six years, named GPT-OSS. This decision marks a return to OpenAI's original mission of ensuring that artificial general intelligence (AGI) benefits all of humanity, while also addressing competitive market dynamics and user demands for privacy, customization, and local AI capabilities.
The GPT-OSS models, gpt-oss-120b and gpt-oss-20b, are state-of-the-art language models designed to run efficiently on consumer hardware. They support local deployment with strong reasoning and tool use capabilities, making them ideal for a wide range of applications.
The larger 120B-parameter model nears the quality of OpenAI’s closed models on core reasoning tasks and supports advanced features like few-shot learning, function calling, and chain-of-thought reasoning. The smaller 20B model is optimized for resource-constrained devices, enabling deployment on PCs and mobile devices, broadening accessibility for developers and users.
NVIDIA has optimized the GPT-OSS models for its RTX GPU lineup, enabling the 120B model to run locally at impressive speeds on certain GPUs, like the RTX 5090. The GPT-OSS-20B model can run entirely on-device, making it possible for smartphones and tablets to perform chain-of-thought reasoning without ever hitting the cloud.
OpenAI's GPT-OSS models are available under the permissive Apache 2.0 licence, allowing for free download, modification, and commercial deployment. They are also available on multiple platforms, including Hugging Face, Ollama, Microsoft AI Foundry Local, and more, making deployment easy for developers.
Databricks is supporting the GPT-OSS models for enterprise-scale workloads, making them accessible for teams running cloud infrastructure. The Ollama app supports local deployment on Windows, macOS, and Linux, and includes SDKs for integration into other applications. Microsoft's AI Foundry Local caters to Windows developers, offering a command line interface and API integration powered by ONNX Runtime.
Moreover, OpenAI's GPT-OSS models can now run on consumer-grade devices, such as smartphones and tablets, using Qualcomm's Snapdragon AI Engine. The models use a mixture-of-experts (MoE) architecture and are some of the first to support MXFP4, a mixed-precision format that delivers high-quality outputs with reduced compute overhead.
However, it's important to note that the GPT-OSS models are text-only, with no built-in multimodal capabilities. They have undergone OpenAI's most rigorous safety testing to date to ensure their reliability and prevent issues like hallucinations, bias, or deceptive behavior.
In essence, OpenAI’s GPT-OSS release revitalizes its open-source ethos, providing powerful AI tools directly to developers while addressing competitive market dynamics and user demands for privacy, customization, and local AI capabilities. This move may reshape AI development by reducing dependence on centralized APIs and fostering a more open AI ecosystem.
Technology has been revolutionized with the introduction of OpenAI's GPT-OSS models, as they are designed to run efficiently on consumer hardware, including smartphones and tablets, thereby broadening the scope of AI applications. These models, such as the GPT-OSS-20B, can even perform chain-of-thought reasoning on-device, demonstrating the potential for decentralized AI capabilities in the future.
