Key Takeaways
1. Tiiny Lab has created the Pocket Lab, a mini PC recognized by Guinness World Records as the “World’s smallest mini PC” for running LLMs with up to 120 billion parameters.
2. The Pocket Lab is compact, measuring 14.2 x 8 x 2.5 cm and weighing only 300 grams, making it portable.
3. Users can easily install popular open-source models with one click and will receive regular OTA software updates.
4. The capability to run LLMs with 120 billion parameters is ambitious, but real-world performance and quantization details are not specified.
5. For average users, a standard mini PC like the Minisforum AI X1 Pro may be a more practical choice and is priced at $1,087.90.
Tiiny Lab is a startup from the US that seems to focus on making hardware for personal LLMs. Their newest creation is a mini PC called the Pocket Lab. This mini PC is very small and has powerful hardware that can run LLMs with up to 120 billion parameters, which is a bold statement.
World’s Smallest Mini PC
According to Tiiny AI, the Pocket Lab mini PC has been recognized as the “World’s smallest mini PC” in the 100B LLM Local category by Guinness World Records. This claim is quite specific, but it does highlight how compact the device is. The Pocket Lab measures just 14.2 x 8 x 2.5 cm (or 5.6 x 3.15 x 1 inches) and weighs only 300 grams, making it very easy to carry around.
Easy Installation and Updates
Tiiny AI also mentions that users can easily install popular open-source models like Llama, Qwen, DeepSeek, Mistral, Phi, and OpenAI GPT-OSS with just one click. They promise regular OTA software updates, which is a good sign. However, we will have to wait and see if they follow through on these promises. On the hardware side, the Pocket Lab mini PC is quite impressive, as shown in the details below:
Tiiny AI asserts that the Pocket Lab can run LLMs with up to 120 billion parameters, but they haven’t specified the quantization levels. Currently, pricing details are not available, but it’s likely that the Pocket Lab mini PC will be on the expensive side. Even with a strong 4-bit quantization, managing a 120 billion parameter model seems unrealistic for a system with only 80 GB of total memory. Real-world performance remains to be seen.
Alternatives for Regular Users
For everyday users who don’t need local LLMs and have average performance needs, a standard mini PC like the Minisforum AI X1 Pro (which is priced at $1,087.90 on Amazon) might be a more suitable option.
Source:
Link