Small Language Model (SLM)

A small language model (SLM) is a lightweight language model typically ranging from 1 billion to 10 billion parameters, compared to the hundreds of billions or trillions found in frontier large language models. Despite their smaller size, SLMs retain core capabilities such as text generation, summarization, translation, and question-answering, often rivaling models 10x their size on domain-specific tasks.

SLMs are optimized for deployment on resource-constrained hardware including smartphones, embedded systems, and edge devices. Their smaller footprint translates to lower latency, reduced memory requirements, and significantly lower inference costs. Organizations often fine-tune SLMs on domain-specific data, producing specialized models that outperform general-purpose LLMs on narrow tasks while running on modest hardware.

Notable examples include Microsoft's Phi-3.5-Mini (3.8B parameters), Google DeepMind's Gemma 3 4B, Meta's Llama 3.2 1B and 3B, Alibaba's Qwen 2.5 1.5B, and Hugging Face's SmolLM2 1.7B. The trend toward capable small models has accelerated as techniques like knowledge distillation, quantization, and architecture improvements continue to close the performance gap with larger models, making on-device AI practical for privacy-sensitive and latency-critical applications.