The Bottom Line: Gemma 4 family with three variants (31B dense, 26B-A4B MoE, E2B compact) is available as a fully managed service on Amazon Bedrock, with native reasoning, function calling, and multimodal support.
Google DeepMind has made the Gemma 4 model family available on Amazon Bedrock. The three open-weight variants offer different sizes and architectures for various requirement profiles, ranging from 2.3 billion effective parameters to 30.7 billion parameters.
The Gemma 4 family from Google DeepMind is now deployed via Amazon Bedrock. It comprises three instruction-tuned variants: Gemma 4 31B (30.7 billion parameters, dense architecture), Gemma 4 26B-A4B (25.2 billion parameters total, 3.8 billion active, Mixture-of-Experts architecture), and Gemma 4 E2B (5.1 billion parameters total, 2.3 billion effective, dense architecture with Progressive Layer Exchange). All variants support text and image input, native function calling for agent-based workflows, and a built-in reasoning mode. The models are available under the Apache 2.0 license.
Independent benchmarks demonstrate the focus on intelligence-per-parameter: Artificial Analysis reports an Intelligence Index of 39 for Gemma 4 31B, significantly above the median of 15 in the weight class 4B–40B for open-weight models. Context windows vary between 128K tokens (E2B) and 256K tokens (31B and 26B-A4B). All variants are available in Standard, Priority, and Flex service tiers.
As a fully managed service on Amazon Bedrock, inference runs exclusively on AWS infrastructure, without requiring you to host model weights or operate inference stacks. This is relevant for CTOs who want to deploy open models in production without compromising data protection, compliance, or control requirements. Your prompts and completions are not used to train models and are not shared with third parties.
The models support over 35 languages out-of-the-box, with pretraining across 140 languages. Since the models are open-weight, you can independently evaluate the architecture and training methodology, benchmark on your own workloads, and perform fine-tuning on proprietary data if needed. Use cases include multimodal agents, lightweight applications, document understanding pipelines, and software engineering workflows.
Source: aws.amazon.com · Published June 15, 2026
Lumi AI News — AI-assisted curation in accordance with Article 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.1.