Amazon Bedrock has announced the availability of the Gemma 4 family of models, developed by Google DeepMind and released under the Apache 2.0 license. These models are designed for a range of deployment scenarios, emphasizing intelligence-per-parameter. The family includes three instruction-tuned variants: Gemma 4 31B, Gemma 4 26B-A4B, and Gemma 4 E2B. These models support reasoning, function calling, and multimodal input across text and image, with benchmarks highlighting their performance.
The Gemma 4 models span from a 2.3B-effective-parameter compact model to a 30.7B-parameter dense model, offering various cost and latency profiles. All variants support built-in reasoning, native function calling, and multimodal input, with over 35 languages supported. The open-weight nature allows independent evaluation and fine-tuning on proprietary data. Access to these models is through a fully managed AWS service, eliminating the need for infrastructure provisioning.
Organizations adopting open-weight foundation models face a trade-off between access to leading models and data protection. Amazon Bedrock removes this trade-off by providing leading open-weight models through a fully managed service with AWS infrastructure and security controls. The models can be used for building multimodal agents, lightweight applications, and document understanding pipelines, with no data shared with third parties.
Source: awsml
