Best Small Language Models in 2026

Microsoft's compact model punches far above its weight on reasoning and math. Runs smoothly on laptops and even mobile devices.

Best for: On-device reasoning and math tasks

Google's mid-sized open-source model with multimodal support. Strong enough for most tasks while fitting on a single consumer GPU.

Best for: Local deployment with multimodal needs

Compact model with excellent multilingual support and function calling. Efficient enough for real-time applications on modest hardware.

Best for: Multilingual edge deployment with tool use

Microsoft's 14B parameter model with exceptional reasoning for its size. Achieves benchmarks competitive with models 3-5x larger.

Best for: Users wanting the best reasoning in a small package

Compact open-source model that runs on consumer hardware. Strong multilingual performance with good general capabilities.

Best for: Resource-constrained deployments needing multilingual support

Meta's compact model with strong general performance and the largest ecosystem of fine-tunes, adapters, and tooling.

Best for: Developers wanting the widest ecosystem of fine-tuned variants