granite-speech-4.1-2b, kanana-2-30b-a3b-instruct, and OpenFold3 models now available on Amazon SageMaker JumpStart

Amazon SageMaker JumpStart now includes three new foundation models: granite-speech-4.1-2b, kanana-2-30b-a3b-instruct, and OpenFold3, broadening the range of AI tools available to AWS customers. granite-speech-4.1-2b supports multilingual automatic speech recognition and bidirectional speech translation across English, French, German, Spanish, Portuguese, and Japanese. The model achieves a word error rate of 5.33% with a real-time factor of approximately 231, making it efficient for enterprise voice workflows. Released under Apache 2.0, it enables scalable transcription, translation, and audio processing for businesses.

Kakao’s kanana-2-30b-a3b-instruct is designed for bilingual Korean-English instruction following and agentic AI workflows. It uses Multi-head Latent Attention and Mixture-of-Experts, activating only 3 billion of its 30 billion parameters per pass for improved throughput. The model supports up to 128,000 tokens via YaRN scaling and is trained with supervised fine-tuning and reinforcement learning to function as a proactive AI collaborator.

OpenFold3, developed by the OpenFold Consortium and the AlQuraishi Lab at Columbia University, predicts all-atom biomolecular complex structures for proteins, DNA, RNA, and small-molecule ligands. Unlike earlier models, it extends structure prediction to multi-chain complexes and heterogeneous biomolecular interactions. This diffusion-based model supports computer-aided drug design and is applicable in academic and pharmaceutical research settings.

Customers can deploy any of these models on AWS via SageMaker JumpStart with minimal setup, either through the SageMaker console or Python SDK. The models are accessible in the SageMaker JumpStart model catalog, allowing users to integrate them into their AI workflows quickly. Documentation is available for guidance on deployment and usage within SageMaker JumpStart.