AWS AI Updates: April 17, 2026

1. Claude Opus 4.7 Lands on Amazon Bedrock

AWS. Anthropic’s newest flagship model is live in Bedrock with high-resolution image support (up to ~3.75 megapixel inputs), sharper agentic coding, and longer-horizon autonomy compared to Opus 4.6. Bedrock serves it through its next-gen inference engine with zero operator data access, keeping prompts and responses invisible to both Anthropic and AWS operators. Pricing matches Opus 4.6 at $5 per million input tokens and $25 per million output tokens. Source

2. Speculative Decoding Hits Trainium2 via vLLM and NeuronX

AWS. AWS published a technical guide on running speculative decoding on Trainium2 using vLLM and the NeuronX Distributed Inference library. Pairing Qwen3-1.7B as the draft model with Qwen3-32B as the target drops inter-token latency from 45 ms to roughly 15 ms for structured prompts and delivers up to 3x throughput speedup on decode-heavy workloads. NxDI supports four speculative decoding modes on Trainium, including fused speculation where draft and target models compile together on Neuron hardware. Source

3. Rede Mater Dei Uses Bedrock AgentCore for Hospital Revenue Cycle Agents

AWS. Brazilian healthcare provider Rede Mater Dei de Saúde documented its production deployment of AI agents on Amazon Bedrock AgentCore to monitor revenue cycle operations across hospitals. The case study covers observability patterns, agent policy enforcement, and telemetry integration — a useful reference implementation for regulated industries deploying agent systems at scale. Source