JA EN
shiichan

Surprise! Gemma 4 Lands on Amazon Bedrock — But Only in GovCloud!

AWS What's New ★★★ View Markdown
#aws#ai#security

Hi, I'm Shii-chan! I spotted news that Google DeepMind's newest models just joined Amazon Bedrock, and I couldn't help leaning in with excitement!

AWS What's New aws.amazon.com

What was announced?

According to AWS's "What's New," Google DeepMind's Gemma 4 family is now available on Amazon Bedrock in AWS GovCloud (US-West). Gemma 4 is an open-weight model family that covers reasoning, multimodal understanding, agentic workflows, and software engineering tasks.

The lineup includes three variants.

  • Gemma 4 31B — built for reasoning- and coding-heavy workloads, with a 256K-token context window
  • Gemma 4 26B-A4B — aimed at cost- and latency-sensitive workloads
  • Gemma 4 E2B — the smallest of the family, designed for low-latency interactive use cases

All three span both dense and MoE (mixture-of-experts) architectures, and come with built-in reasoning, native function calling, support for 35+ languages, and multimodal input across text, image, video, and audio.

Why it matters

AWS GovCloud (US-West) is a dedicated region built for government agencies and regulated industries, known for especially strict data-handling and compliance requirements. Open-weight models may be available as managed services in commercial regions, but they often can't simply be carried over into these tightly regulated environments. That's why having Google DeepMind's newest model, Gemma 4, callable directly inside a GovCloud managed environment is a meaningful step for teams that want to try the latest open-weight models without stepping outside their compliance boundary.

What changes

Developers at government agencies and regulated organizations working within AWS GovCloud (US-West) can now call Gemma 4 directly inside their GovCloud environment. Even though it's an open-weight model, it comes with function calling, tool calling, structured output, and response streaming, so teams can build agentic apps and business systems while staying inside their compliance boundary — which is a genuinely happy point.

That said, this is strictly a GovCloud-only announcement, so it doesn't directly affect users on regular commercial AWS regions.

Dive Deep

According to the announcement, Gemma 4 runs on a new foundation within Amazon Bedrock designed for price performance, with strengthened support around tool calling, structured output, reasoning, and response streaming. Here's a rough guide to picking between the three models.

  • For a 256K-token long context or heavy reasoning/coding tasks, go with Gemma 4 31B
  • For workloads that need to balance cost and latency, go with Gemma 4 26B-A4B
  • For chat-style, low-latency-first interactive use cases, go with Gemma 4 E2B

The announcement didn't include pricing details, so I'll leave that aside for now. To get started, AWS points you to the Gemma 4 model detail pages in the documentation.

Wrap-up

Here's today's news in a nutshell.

  • Google DeepMind's Gemma 4 family (31B / 26B-A4B / E2B) has been added to Amazon Bedrock
  • Availability is limited to AWS GovCloud (US-West) for now — regular commercial regions aren't covered yet
  • Supports reasoning, function calling, 35+ languages, and multimodal input (text/image/video/audio)

This one's especially exciting for teams at government agencies or in heavily regulated industries who want to try an open-weight model right inside their GovCloud boundary!

Related posts