As AI Models are getting more and more powerful—even open-source models can deliver impressive performance and handle most task requirements, which is why many people want to deploy them locally. However, deploying large-parameter models requires expensive hardware that not everyone can afford. Recently, foreign AI startup Tiiny AI released a product claiming to solve this pain point, and it’s also billed as the world’s smallest super…Computer“Tiiny AI Pocket Lab” can deploy up to 120B parameters locally. LLM Large language model，no internet required.
Touted as the world’s smallest supercomputer, the Tiiny AI Pocket Lab has made its debut! Despite being pocket-sized, it can run 120B AI models locally.
Tiiny AI is an American AI startup, and this time it has unveiled the world’s smallest personal AI supercomputer, the “Tiiny AI Pocket Lab.” Before we get into the specs, there’s one certification worth mentioning first.
According to a report from TechPowerUP, this supercomputer has officially received Guinness World Record certification for “World’s Smallest MiniPC (capable of running 100B LLM locally).” I’m sure many people have never heard of this certification—neither had I. It seems this is the first time it’s been publicly unveiled.
The Tiiny AI Pocket Lab is very compact, measuring only 14.2 x 8 x 2.53 cm, roughly the same size as a common 10,00mAh power bank, and weighs just 300 grams, making it easy to slip into a pocket and carry around. Tiiny AI officially claims that this device can successfully deploy a model with 120 billion parameters (120B) without requiring an internet connection, server, or high-end GPU, making it well-suited for “PhD-level reasoning, multi-step analysis, and deep contextual understanding.”
It’s truly incredible that such a small device can run a large language model.
The Tiiny AI Pocket Lab is also very power-efficient, operating within a power consumption range of only 65 W, far lower than the energy and carbon emissions of traditional GPU systems.
Tiiny AI’s GTM Director Samar Bhoj said:
雲端 AI 帶來了巨大進展，但同時也造成依賴、脆弱性與永續性挑戰。透過 Tiiny AI Pocket Lab，我們相信智慧不應屬於資料中心，而應屬於每一個人。這是把大型模型能力從雲端帶到每一部個人裝置，使其真正可及、隱私與個人的第一步。
According to public information from Tiiny AI, Tiiny AI Pocket Lab supports one-click installation of popular models such as GPT-OSS, Llama, Qwen, DeepSeek, Mistral, and Phi, and can quickly deploy common open-source AI agents such as OpenManus, ComfyUI, Flowise, Presenton, Libra, Bella, and SillyTavern.
You might be wondering why it can deploy large language models? The key is its dedicated NPU, which delivers 190 TOPS of computing power, along with 80GB of built-in LPDDR5X memory.
Moreover, the official announcement also mentioned that the core of Tiiny AI Pocket Lab lies in two major technological breakthroughs, which also allow large language models to run on extremely small devices.
- TurboSparseTechnology that makes AI operate more efficiently. It’s like having the brain engage only the neurons needed at the moment, rather than straining to run at full speed every time. This allows AI to compute much faster while staying just as smart.
- PowerInferAn open-source engine with over 8,000 stars on GitHub. It intelligently coordinates the CPU and NPU in your computer to work together on demanding AI tasks. The result is that you can enjoy powerful performance comparable to large servers while consuming very little electricity.
Don’t get too excited just yet, because this product hasn’t been officially launched. It’s expected to be showcased at CES 2026 next year. As for whether it’s really that powerful, we’ll have to wait until we see the actual product. The release date and pricing are also unclear.
Tiiny AI Pocket Lab Specifications:
| Classification | Specification |
|---|---|
| Processor | ARMv9.2 12-core CPU |
| AI computing power | In-house heterogeneous architecture module (SoC + dNPU), approximately 190 TOPS |
| Memory and Storage | 80GB LPDDR5X RAM + 1TB SSD |
| Model capacity | Fully on-device, runs LLMs with up to 120B parameters |
| Power efficiency | 30W TDP, with typical system power consumption around 65W. |
| Dimensions and Weight | 14.2 × 8 × 2.53 cm, approximately 300 g |
| Ecosystem | One-click deployment of multiple open-source LLMs and Agent Frameworks |
| Connectivity | Can operate completely offline — no network or cloud required |
Source: KOCPC Chinese