Ultra-compact Alibaba neural network for smartphones
Alibaba has introduced the compact MNN-Pocket-0.3B language model, weighing just 1.73 MB and capable of running on smartphones.
The neural network contains 300 million parameters and demonstrates high performance: on an iPhone 18 Pro, it generates up to 350 tokens per second.
In test runs, the model outperformed the larger Qwen3-0.6B in almost all tasks, effectively handling code, text, and mathematical calculations.
The AI's energy efficiency is at a high level: processing 100 requests on a OnePlus 13 smartphone consumes only 0.3% of the battery charge.
The model is available for free download, fine-tuning, and integration into third-party applications under an open-source license.










