AWS News Weekly

AI & Bedrock Nova 2.5 Sonic: GA voice model, better reasoning/tool calls, lower latency Claude Haiku 5.5: ~75% cheaper than Haiku 4.5, effort controls (+GovCloud) GLM 5.3: 753B MoE, 1M context, 128K output GPT-6.1 Sol: Ultrafast mode; reasoning summaries for OpenAI models TwelveLabs Pegasus 1.5: Video-to-text with time-coded metadata Kiro GovCloud: Opus 5.5 (half the tokens) & Sonnet 5.5 (30% faster) Strands Bidi Agents: GA for production voice agents Security & Compliance GuardDuty: RDS

This week: Amazon Bedrock adds Claude Sonnet five point five plus new Grok and OpenAI models, Aurora PostgreSQL can now query Iceberg and Parquet data in S three with no ETL, and EKS gets Kubernetes one point thirty-seven with scale-to-zero autoscaling. Plus Security Hub remediation plans that fix multiple exposures at once, smarter S three Vectors filtering, and service retirements to plan for. AI & ML Bedrock: Claude Sonnet 5.5 (incl. GovCloud), Grok 4.7, GPT-6.1 Sol GPT-6 Astra: UltraFast m

This week: Claude Opus five point five lands in Amazon Bedrock alongside OpenAI's new GPT six Sol and Luna models, CloudWatch Omni brings AI-first observability built on OpenTelemetry, and EventBridge relaunches custom event buses with strict ordering and cross-account sharing. Plus DynamoDB multi-region strong consistency expands to fifteen regions, and EMR gets Long Term Support releases. AI & ML Bedrock: Claude Opus 5.5, fewer tokens, cheaper than Opus 5; in GovCloud Bedrock: OpenAI GPT-6 S

This week: EC2 launches application status checks for monitoring app-level health directly alongside instance checks, IAM introduces Account Access Manager for assigning roles to Identity Center users, and S3 now shows the specific policy ARN in access denied errors for faster troubleshooting. Plus new SageMaker JumpStart models, Bedrock cost allocation by IAM principal, and QuickSight adds deny-by-default for AI capabilities. Compute & Infrastructure EC2: App status checks monitor web servers/

This week: DynamoDB adds native vector search with single-digit millisecond latency, Amazon Bedrock brings OpenAI GPT models with million-token context windows, and Aurora Serverless now scales to 12 ACUs in under a second. Plus ECR supports 200GB image layers, Lambda gets 3Gbps bandwidth, and Bedrock AgentCore launches in GovCloud. Databases DynamoDB: Native vector search, single-digit ms latency, 99%+ recall, trillions of vectors Aurora Serverless: Scales to 12 ACUs in <1 second, up to 256 A

This week: AWS launches Amazon Interconnect for purpose-built multicloud connectivity with Oracle and Google Cloud. MSK Express brokers now deliver Kafka data directly to S3 and Iceberg tables, cutting costs up to 60 percent. Transit Gateway adds policy-based routing for traffic steering by source, destination, and protocol. Plus IAM Identity Center gets multi-region replication and Bedrock drops GPT 5.6 prices up to 80 percent. Database RDS SQL Server: Restore TDE-encrypted DBs directly to Mul