Gemma 3 27B IT (Amazon Bedrock, US)

Gemma 3 27B IT is Google's open-weights contribution to the Gemma 3 family, designed around a 27-billion parameter architecture that brings true multimodality to the lineup. It accepts combined image and text prompts and produces text-only responses, letting teams build vision-language assistants, document and screenshot analyzers, and chat applications that need to reason over both visuals and prose. A 131,072-token context window supports long conversations, lengthy document ingestion, and multi-image workflows, while built-in understanding of more than 140 languages broadens its reach for global products and multilingual content pipelines.

In practice, the model is positioned as a balanced mid-size option that trades raw scale for efficiency and accessibility. Reported benchmark results highlight dependable behavior, including perfect scores on ethics evaluations, strong general-knowledge accuracy around 98 percent, and solid coding performance near 89 percent, with mathematics in a moderate range around 84 percent. Because the weights are openly published, developers can self-host, fine-tune, or route the model through hosted services, making it a practical fit for teams that want vision-language capability, long context, and multilingual coverage without committing to the largest frontier models.