LiberaGPT supports the largest and most intelligent AI model to run offline on an unmodified iPhone
Independent British software house 5N6 has unveiled a major update to LiberaGPT, its recently launched iPhone app, adding support for a record-breaking 24 billion-parameter large language model that runs entirely offline on the latest generation of iPhone devices, with complete privacy and no requirement for cloud data centres or additional hardware. Until recently, the idea of running a model of this size privately on a smartphone would have been regarded as technically out of reach for current flagship handsets. LiberaGPT now brings that capability to iPhone, ensuring that prompts and responses remain entirely on the device. The comparison with earlier generations of AI highlights the scale of that advance. OpenAI’s GPT-2, introduced in 2019, was widely recognised as a landmark model, yet it operated at 1.5 billion parameters and depended heavily on centralised infrastructure. Since then, major gains in efficiency and model design have reshaped expectations, allowing more powerful systems to operate without the same computational burden. Alongside the new 24 billion-parameter model, the latest LiberaGPT release also includes a range of smaller models for users who prefer faster performance and lower resource use. Together, these options give users the flexibility to choose the balance of speed and capability that suits them best, while keeping the entire experience private and on-device. What matters here is not only the size of the model, but the fact that it runs at all on consumer mobile hardware. Models in this class have usually belonged to server infrastructure or desktop systems. This small independent software house has brought that level of capability to the iPhone through careful optimisation and precise memory management. Capable local GPU processing gives users more privacy, faster response times, and the ability to use advanced AI models without dependence on a network connection or contribution to AI data centre infrastructure […]