Google DeepMind's Gemma 4 family of open-weight models is now available on Amazon Bedrock via the bedrock-mantle endpoint. The family includes three instruction-tuned variants: Gemma 4 31B (dense, 256K context), Gemma 4 26B-A4B (mixture-of-experts, 4B-class inference cost), and Gemma 4 E2B (compact, 128K context). All variants support built-in reasoning mode, native function calling, multimodal text and image input, and 35+ languages. Access is through an OpenAI-compatible API, making migration from existing OpenAI SDK integrations straightforward. The post covers service tiers (Standard, Priority, Flex), on-demand scaling best practices including traffic ramping procedures, exponential backoff for 503 errors, implicit prompt caching, and IAM setup requirements.
