Introducing Gemma 4 models on Amazon Bedrock
TL;DR
AWS is adding Google DeepMind’s Gemma 4 family to Amazon Bedrock. The lineup includes three instruction-tuned models: Gemma 4 31B, Gemma 4 26B-A4B, and Gemma 4 E2B, released under Apache 2.0. The models cover dense and mixture-of-experts designs. 26B-A4B has 25.2B total parameters but activates 3.8B per token. 31B and 26B-A4B support 256K context; E2B supports 128K.
Nauti's Take
AWS turns an open model into something enterprises can actually buy, govern, and run inside familiar infrastructure. That is the practical angle: Gemma 4 does not need to beat every frontier model to matter if procurement, IAM, API access, and operations fit the AWS lane.
The catch is the launch-story bias. Teams should test real images, long documents, tool calls, latency, and cost per completed task before treating the benchmark narrative as a deployment plan.
Briefingshow
For teams already on AWS, Gemma 4 becomes much easier to evaluate: no self-hosting, no custom inference stack, while keeping open weights and an Apache 2.0 license. The useful mix is long context, image input, and tool calling in one managed service. The real test is cost, latency, and quality in actual agent and document workflows.
Sources
