AWS: Google Gemma 4 Models Available on Amazon Bedrock — Three Variants Under Apache 2.0 License

Amazon Bedrock has added three models from Google’s Gemma 4 family of open models, making them available through managed AWS infrastructure under the Apache 2.0 license.

Which three Gemma 4 variants are coming to Bedrock?

AWS offers three variants: Gemma 4 31B (dense model with 256K context), Gemma 4 26B-A4B (MoE model with 3.8 billion active parameters and 256K context), and the compact E2B (5.1 billion parameters, 128K context). MoE (Mixture of Experts) is an architecture in which only a subset of parameters is activated for each token, allowing the model to retain the capacity of a large model while keeping inference costs lower.

How capable is Gemma 4 31B?

Gemma 4 31B achieves an Intelligence Index of 39 as measured by Artificial Analysis, versus a median of 15 for the class of open models from 4 to 40 billion parameters. The models were trained on more than 140 languages and support 35+ languages in use, with native function calling, a built-in reasoning mode, and multimodality (text and images).

How are the models used on AWS?

Gemma 4 on Bedrock is available through an OpenAI-compatible API via the bedrock-mantle endpoint, which simplifies migrating existing integrations. AWS offers three pricing tiers: Standard, Priority (up to 25% better output tokens per second compared to Standard), and Flex (discount for latency-tolerant workloads). The models are initially available in four regions — two US East, US West, and EU Frankfurt.

Why does the Bedrock arrival matter?

The Apache 2.0 license permits commercial use and modification, so bringing Gemma 4 models to Bedrock combines an open model with managed hosting. Companies gain a powerful open model without needing their own GPU infrastructure, with the ability to choose a pricing tier to match their latency requirements.