Google Releases Gemma 3 Open Weights Models with 27B Parameter Version Topping Open-Source Benchmarks

Google DeepMind has released Gemma 3, the third generation of its open-weights language model series, with a 27B parameter version that leads open-source benchmarks across reasoning, coding, and multilingual tasks. Unlike Llama 3 70B which requires multi-GPU infrastructure, Gemma 3 27B runs on a single high-end GPU — making it accessible for APAC enterprises with on-premises deployment requirements who cannot provision the infrastructure for larger open-source models.

Gemma 3's multilingual capabilities are particularly relevant for APAC deployment: the model includes significantly expanded training data in Japanese, Korean, Simplified and Traditional Chinese, Indonesian, Vietnamese, and Thai — addressing the gap in smaller models that perform strongly in English but degrade substantially in Asian language tasks. APAC enterprises deploying Gemma 3 for internal applications (document analysis, customer query handling, content generation) report stronger performance on APAC-language inputs than previous Gemma versions.

For APAC regulated industries (financial services, healthcare, government) with data sovereignty requirements that preclude cloud-hosted API models, Gemma 3 27B provides a genuinely capable on-premises alternative that can run on infrastructure available to mid-market organisations — not only large enterprises with GPU clusters. The single-GPU inference requirement means APAC enterprises can begin Gemma 3 deployment on existing ML infrastructure without capital investment in additional compute.