Smaller AI Models Pack Surprising Power, Rivaling Larger Counterparts

The AI industry is witnessing a shift away from the 'bigger is better' mantra as small language models (SLMs) demonstrate remarkable performance at a fraction of the cost. Recent examples include Microsoft's Phi-3-mini, which despite its compact size has reportedly outperformed some larger models in language, coding, and math benchmarks. Meanwhile, Apple's 2025 on-device foundation model, with just 3 billion parameters, leverages optimizations like KV-cache sharing and 2-bit quantization-aware training to run privately on Apple silicon. These developments suggest the frontier is moving toward smaller, specialized, local, and fast AI that can be deployed everywhere.