SAN FRANCISCO: OpenAI, the developer of ChatGPT, has launched two new artificial intelligence (AI) models that users can customise and operate locally on personal devices, marking a shift towards more open-access AI technology.
Named GPT-oss-120b and GPT-oss-20b, these models are “open-weight language models”, meaning their internal parameters are accessible. Unlike the closed architecture of ChatGPT, these models are designed for flexibility, allowing businesses and developers to fine-tune them to suit specific tasks and operate independently from cloud services.
The more powerful GPT-oss-120b can run on devices with high-end graphics cards, while the smaller GPT-oss-20b is optimised for laptops with at least 16GB of RAM. This makes advanced AI capabilities available to users without the need for massive computing infrastructure.
Although OpenAI did not disclose the training data used for these models, both are capable of generating text, writing code, and performing online research — similar to existing AI models but with the added benefit of local control and customisation.
This launch is a strategic move to serve organisations that prefer on-premises AI deployments, particularly those with privacy concerns or specialised operational needs. Customisable models like these are ideal for sectors such as healthcare, finance, or education, where data security and model transparency are critical.
The move also places OpenAI in closer competition with companies offering open-source or open-weight alternatives, as it expands its reach to independent developers and enterprises seeking more autonomy over their AI tools.
With growing demand for decentralised and adjustable AI systems, OpenAI’s new models cater to a broader range of users and devices, supporting more personalised and secure applications in AI development.
