US deep-tech startup Tiiny AI Inc has introduced Tiiny AI Pocket Lab, a pocket-sized personal Artificial Intelligence supercomputer that the company says can run up to a full 120-billion-parameter large language model entirely on-device. The device has been officially verified by Guinness World Records in the category “The Smallest MiniPC (100 LLM Locally)”, highlighting its focus on local model execution instead of cloud-based processing. Tiiny AI positions the Pocket Lab as a way to bring large-scale Artificial Intelligence to the edge, making advanced systems more personal, accessible and integrated into daily life.
Tiiny AI Pocket Lab is described as a compact inference system that can run an LLM with as many as 120 billion parameters. The unit measures about 14.2 x 8 x 2.53 cm, weighs around 300 grams, and operates within a 65W power envelope, aiming to deliver large-model performance at significantly lower energy consumption than conventional GPU-backed Artificial Intelligence setups. The hardware is built around an ARMv9.2 12-core CPU and a dedicated neural processing unit capable of delivering about 190 TOPS of Artificial Intelligence compute, backed by 80GB of LPDDR5X memory and 1TB of storage. According to the company, Tiiny AI Pocket Lab operates in the “golden zone” of personal Artificial Intelligence (10B-100B parameters), which it claims is ideal for more than 80 per cent of real-world needs and reportedly offers intelligence comparable to GPT-4o, with PhD-level reasoning, multi-step analysis and deep contextual understanding.
The system relies on two core technologies, TurboSparse and PowerInfer, to make large-parameter models viable on such a small device. TurboSparse applies neuron-level sparse activation to improve inference efficiency, while PowerInfer is an open-source inference engine that accelerates heavy LLM workloads by dynamically sharing computation between the CPU and NPU, enabling performance previously associated with professional GPUs worth thousands of dollars. Tiiny AI Pocket Lab supports one-click installation of open-source models including OpenAI GPT-OSS, Qwen, DeepSeek, Llama, Phi and Mistral, and can deploy Artificial Intelligence agents like OpenManus, ComfyUI, Flowise, Libra, Presenton, Bella and SillyTavern, with the company promising continuous updates and official OTA hardware upgrades, and stating that these features will be released at CES in January 2026. By reducing dependence on cloud servers, Tiiny AI Pocket Lab is framed as a way to cut operational costs, mitigate latency and sustainability concerns associated with data centres, and deliver advanced Artificial Intelligence capabilities to individuals, especially in constrained environments, building on a founding team formed in 2024 with engineers from institutions such as MIT, Stanford, Intel, Meta, HKUST and SJTU.
