Google's Gemma 4 Open-Weight Models Now Available on Amazon Bedrock
Summary
- • AWS adds Google DeepMind's Gemma 4 family—31B dense, 26B-A4B MoE, and E2B—to Amazon Bedrock under Apache 2.0 license
- • All variants support built-in reasoning, native function calling, multimodal text+image input, and 35+ languages
- • Gemma 4 31B scores an Intelligence Index of 39 vs. a median of 15 in the 4B–40B open-weight class per Artificial Analysis
- • Customer prompts and completions are not used to train Google's models; data remains within AWS infrastructure
Details
Gemma 4 on Amazon Bedrock
Google DeepMind's Gemma 4 family now available as a fully managed service on AWS Bedrock, covering three model variants for different cost and latency profiles
MoE architecture (26B-A4B)
Mixture-of-Experts model activates only 4B of 26B total parameters per request, cutting inference costs while maintaining output quality
Intelligence Index 39 vs. median 15
Gemma 4 31B scores more than double the median benchmark score in the 4B–40B open-weights class per Artificial Analysis
Fully managed, no self-hosting
AWS handles all inference infrastructure; organizations do not need to provision hardware, host model weights, or operate inference stacks themselves
Apache 2.0 + 140+ language pre-training
Open-weight license permits proprietary fine-tuning; models pre-trained across 140+ languages with runtime support for 35+
Key technical and commercial details of Gemma 4's availability on Amazon Bedrock
What This Means
The arrival of Gemma 4 on Amazon Bedrock gives enterprise teams a practical path to Google's top-performing open-weight models within a fully managed, compliance-ready AWS environment — eliminating the operational overhead of self-hosting. Gemma 4's strong benchmark performance relative to its size class, combined with MoE efficiency and Apache 2.0 licensing, positions it as a compelling alternative to proprietary API-only models for production workloads. This move signals deepening integration between Google DeepMind's open-weight model program and major cloud providers, intensifying competition in the managed open-model space. Organizations building multimodal agents, document pipelines, or multilingual applications now have a turnkey enterprise option.