Gemma 3

Overview

Google's open-weight 27B parameter model designed to run on a single GPU or even consumer hardware. Gemma 3 delivers surprisingly strong performance for its size, with multimodal vision capability and 128K context.

Why it matters

Gemma 3 is Google's strategic play to own the on-device and edge AI tier. A 27B model that fits on a single GPU and punches well above its weight class is exactly what developers need for local inference, embedded applications, and privacy-sensitive deployments. For startups building products that need to run without cloud API calls — think on-device assistants, offline tools, or latency-critical applications — Gemma 3 is often the best option. Its permissive licensing and Google-backed optimization for diverse hardware make it a pragmatic choice.