FinInsights deployed a JumpStart foundation model endpoint for low-latency financial-text generation on an ml.g5.xlarge instance. After a successful pilot, they need 3× higher sustained throughput with similar latency tail behavior. What is the most direct, supported change to the SageMaker JumpStart endpoint to meet this requirement?

The correct option directly addresses the need for 3x higher sustained throughput with similar latency. Increasing the instance count (scaling out) or using a larger instance type (scaling up) are standard methods to improve throughput for inference endpoints. Enabling autoscaling ensures the endpoint can dynamically adjust to sustained load. Redeploying with the new configuration applies these changes. The other options are incorrect because: HyperParameterTuningJob is for training, not for optimizing deployed inference throughput. Elastic Inference accelerators are for cost-effective inference for specific deep learning models, but they don't typically provide a 3x throughput increase for foundation models and are not the most direct solution for this scale. SageMaker Serverless Inference is designed for intermittent traffic patterns and can have higher cold start latencies, which might not meet the "similar latency tail behavior" requirement for sustained high throughput.