AWS AI Updates: April 30, 2026

1. Gemma 4 Lands on SageMaker JumpStart in E4B, 26B-A4B, and 31B Variants

AWS. SageMaker JumpStart now hosts three Google DeepMind Gemma 4 instruction-tuned models: E4B, 26B-A4B, and 31B. All three offer multimodal input, configurable reasoning, native function calling for agentic workflows, and coverage of more than 140 languages, with the E4B variant additionally accepting audio input for speech recognition and cross-language translation. Customers can deploy directly through SageMaker Studio or the Python SDK without leaving the JumpStart catalog. Source

2. SageMaker JumpStart Adds Multilingual Embedding, Table Detection, and Polish-Tuned LLM

AWS. A second JumpStart drop expands the catalog with three specialized models: Paraphrase-multilingual-MiniLM-L12-v2 for semantic similarity across more than 50 languages (cross-lingual search and clustering), Microsoft’s DETR-based Table Transformer for detecting tables in PDFs and scanned documents, and Bielik-11B-v3.0-Instruct, an 11B model tuned for dialogue and reasoning across 32 European languages with a Polish and enterprise focus. The mix targets retrieval, document AI, and regional-language deployments that don’t fit the dominant English-centric foundation models. Source