OpenAI goes 'open' with its new gpt-oss LLMs

OpenAI dropped two powerful open-weight AI models, its first free release since GPT-2 in 2019. The gpt-oss-120b and gpt-oss-20b models are freely downloadable, fully customizable, and powerful enough to rival the company's premium offerings, while running entirely on your own hardware.The larger gpt-oss-120b packs 117 billion parameters but uses smart mixture-of-experts technology to activate just 5.1 billion per task, letting it run efficiently on a single 80GB GPU. Its smaller sibling, gpt-oss-20b, is the real game-changer, it runs comfortably on any laptop with 16GB of RAM while delivering performance comparable to OpenAI's o3-mini model.Both models shine at reasoning tasks, can browse the web, execute Python code, and even connect to OpenAI's cloud models when they need extra capabilities like image processing. They're essentially mini versions of the company's flagship reasoning models, complete with adjustable chain-of-thought processing that you can dial up or down depending on your needs.The models are designed to be flexible about where they run, your laptop, company servers, or cloud platforms like Azure and AWS. OpenAI worked with Microsoft to get optimised versions running on Windows devices, and they're already live on Hugging Face and other major platforms.Unlike fully open-source releases, these models only provide weights, the learned parameters that enable functionality, while keeping training data and source code proprietary. However, the models support full chain-of-thought reasoning, tool use like web search and Python execution, and can seamlessly integrate with OpenAI's cloud-based services for tasks requiring multimodal capabilities. OpenAI also put these through some serious safety testing, even trying to deliberately break them to see what bad actors might pull off. Under the Apache 2.0 license, developers can modify them however they want.