Google's Gemma 4 lands on AWS GovCloud via Bedrock

Government and public-sector developers building on AWS now have direct access to Google DeepMind's Gemma 4 model family, which Amazon has added to Bedrock within its GovCloud (US-West) region. The move brings an open-weight, multimodal model line into an environment reserved for regulated and sensitive workloads.

Gemma 4 ships in three configurations rather than a single model. Gemma 4 31B is the largest, built for reasoning and coding-heavy tasks with a 256K-token context window that lets it handle long documents or codebases in a single pass.

Gemma 4 26B-A4B is aimed squarely at teams optimizing for cost and latency, while Gemma 4 E2B is the smallest of the three, intended for low-latency, interactive applications where response speed matters more than raw capability. Together, the lineup spans dense and mixture-of-experts (MoE) architectures, giving developers a choice between consistent per-token compute and the sparser activation patterns MoE designs allow.

What the models can do

Across all three sizes, Gemma 4 supports more than 35 languages and accepts multimodal input, including text, images, video, and audio. Google has also built in native function calling and reasoning capabilities, which reduces the amount of custom scaffolding developers previously needed to wire these models into agentic workflows or tool-use pipelines.

The GovCloud availability matters less for the model's underlying capabilities than for where it can now legally and operationally run. GovCloud is AWS's isolated region built to meet US government compliance regimes such as ITAR and FedRAMP, and adding Gemma 4 there means public-sector teams, defense contractors, and regulated enterprises can use the model without routing data through AWS's commercial regions.

That distinction is likely to matter more to procurement officers than to typical developers. For most commercial Bedrock users, Gemma 4's arrival is simply another open-weight option alongside existing foundation models, differentiated mainly by its long context window on the 31B variant and its explicit tiering by latency and cost.

No pricing details specific to the GovCloud deployment were included in AWS's announcement, and the facts available don't specify whether Gemma 4 is also rolling out simultaneously to Bedrock's standard commercial regions. That leaves open whether this GovCloud listing is a coordinated global launch or a compliance-driven follow-up to a broader release.