Claude Sonnet 5 lands on Bedrock alongside Graviton5 instances

AWS's weekly roundup centered on making Anthropic's newest Sonnet available where enterprises already run: Claude Sonnet 5 is now on Amazon Bedrock as AWS's most capable Sonnet model for coding and agentic tasks. Sonnet 5 launched July 1 at aggressive $2/$10 per-million-token pricing with browser use, terminal access and autonomous planning, and putting it on Bedrock gives regulated customers a supported, governed path to it.

AWS paired the model news with infrastructure: new EC2 C9g and C9gd instances built on Graviton5 deliver roughly 25% better compute performance and a 5x larger cache than Graviton4, targeting the CPU-side of AI serving and general compute. SageMaker Inference now supports container image caching for up to 2x faster scale-out — a direct answer to cold-start latency in agentic workloads — and OpenSearch added a log-analytics engine claiming up to 4x better price-performance.

The governance angle is increasingly central. AWS also showed how to enforce zero data retention on Bedrock using Bedrock Projects and service control policies, explicitly framed around new models that require third-party data sharing — a nod to enterprises nervous about where prompts and outputs go, and a competitive contrast with the Alibaba/Claude-Code back-door anxieties dominating the week.

Competitively, AWS's pitch is neutrality plus control: host every major model (Claude via Bedrock, Claude in Microsoft Foundry is the Azure equivalent) with enterprise guardrails. The developer community's warm reception of Sonnet 5's benchmarks (63.2% SWE-bench Pro, 80.4% Terminal-Bench) at Sonnet 4.6 pricing gives AWS a genuinely in-demand model to sell. Watch whether Graviton5's cache gains translate to real inference-cost wins for customers.