Amazon Web Services has made public the availability of Google DeepMind’s Gemma 4 models on Amazon SageMaker JumpStart. With this new addition, Amazon Web Services will be providing its customers with three highly optimized and instruction fine-tuned models – Gemma 4 E4B, Gemma 4 26B-A4B, and Gemma 4 31B, which provide advanced multimodal abilities.
With growing demands from enterprises regarding AI systems that can handle different forms of input data, Gemma 4 offers highly sophisticated capabilities, such as reasoning and function calling, while supporting multiple languages, exceeding 140.
Enhanced Intelligence and Versatile Architectures
The Gemma 4 family, built on the same research and technology as Google’s Gemini models, is designed to deliver industry-leading “intelligence-per-parameter.” By making these models available in SageMaker JumpStart, AWS allows customers to deploy and scale sophisticated AI workflows on secure, managed infrastructure with just a few clicks.
Each of these three new models is equipped with a common set of features designed specifically for today’s enterprise needs:
- Reasoning Built-In: The presence of a “Thinking” feature means that each model can analyze data progressively before providing a response.
- Multimodal Intelligence: Support for natural text and image processing in combination with video interpretation through frame-by-frame sequence.
- Autonomous Functionality: Native function calling makes these models capable of working with third-party APIs on their own.
- Document Interpretation: Advanced capabilities in OCR, handwriting detection, and parsing of sophisticated PDF documents, charts, and GUIs.
Also Read: ClickHouse and Google Cloud Unveil Deepened Strategic Alliance to Redefine Real-Time Analytics and Cloud Flexibility
Model Selection Based on Workload
Today, there are several options for choosing the exact model to meet your computing requirements:
- Gemma 4 E4B: An effective model focused on fast processing. What should be highlighted is that the E4B model offers audio capabilities, allowing automatic speech recognition (ASR). This allows users to conduct speech-to-text translation.
- Gemma 4 26B-A4B: This model operates on the principle of Mixture-of-Experts, combining advanced reasoning with efficiency.
- Gemma 4 31B: This is a dense and powerful model that is capable of frontier-level intelligence in coding, reasoning, and multilingualism.
Streamlined Deployment via Amazon SageMaker
The inclusion of Gemma 4 as an option in SageMaker JumpStart has made deployment of open weight models easy. Customers will now be able to deploy these models from the SageMaker Studio UI or even the SageMaker Python SDK. This ensures that the models are deployed in a secure AWS environment.
For those looking to start their experience with Gemma 4, developers can check out the “Models” tab in SageMaker Studio for detailed Gemma 4 cards.
