Can powerful AI models actually run on your phone?

The answer

Small language models (SLMs) are transforming on-device AI. Microsoft's Phi-3 Mini, released in April 2024, packs 3.8 billion parameters yet outperforms models twice its size on reasoning benchmarks. Google's Gemma 2, launched in June 2024, comes in 2B, 9B, and 27B variants optimized for laptops and edge devices. Meta's Llama 3.2, released in September 2024, introduced 1B and 3B text models specifically designed for mobile deployment - the 3B model outperforms Gemma 2 2.6B and Phi-3.5 Mini on instruction-following and summarization tasks. These SLMs enable AI features without cloud connectivity, offering faster responses, better privacy, and lower costs, representing a fundamental shift from the bigger-is-always-better paradigm.