Amazon SageMaker JumpStart adds 100+ models in first three quarters of 2026 and simplifies deployment

From January to September 2026, SageMaker JumpStart added over 100 models covering 30 providers. Alibaba Qwen was the largest contributor with 29 models spanning reasoning, code generation, speech recognition, text-to-speech, embeddings, and reranking; NVIDIA contributed 17, including the Cosmos3 world foundation model and the Nemotron series; Mistral, Google Gemma 4, IBM Granite 4.0, Cohere Command A, and others are also included. On the deployment side, four new named configurations were added: balanced, throughput-first, latency-first, and cost-first. Throughput-first uses speculative decoding, latency-first minimizes time to first token, and Neuron-optimized configurations for Inferentia and Trainium are included. The new Model Hub places the full catalog on the SageMaker AI console homepage, with filtering by modality, publisher, model size, license, and framework. For teams deploying open-source models on AWS, Hugging Face deep links jump directly to a preconfigured Studio environment, removing the manual prerequisite steps of configuring IAM roles and permissions.