Amazon Bedrock 在中東地區（阿聯酋與巴林）推出 Anthropic Claude 模型全球跨區域推論

🔑 增強重點摘要

• •Global cross-Region inference offers approximately 10% savings on input and output token pricing for Claude Sonnet 4.5 compared to geographic profiles, with pricing based on the source Region.[5]
• •Two inference profile types exist: Geographic CRIS (routes within geographies like US, EU, APAC) for data residency compliance, and Global CRIS (routes worldwide to over 20 commercial Regions) for maximum scale.[2][3]
• •Feature supports on-demand inference, batch inference, agents, model evaluation, prompt management, and prompt flows beyond standard invocation.[5]
• •Similar rollout occurred in Thailand, Malaysia, Singapore, Indonesia, and Taiwan for Claude Opus 4.6, Sonnet 4.6, and Haiku 4.5, emphasizing cost efficiency and routing across APAC.[2]

🛠️ 技術深入

• •Uses inference profile IDs like global.anthropic.claude-sonnet-4-5-20250929-v1:0; requests from source Region route to destination Regions over AWS managed network with end-to-end encryption.[5]
• •Data at rest (logs, knowledge bases, configurations) stays in source Region; only transient prompts/results move encrypted during inference.[2]
• •ARN pattern for global profile: arn:aws:bedrock:REGION:ACCOUNT:inference-profile/global.MODEL-NAME; requires IAM permissions for source and potential destination Regions.[5]
• •Throughput hierarchy: Global CRIS > Geographic CRIS > single-Region inference; quotas detailed in AWS service limits for requests/tokens per minute.[3][4]

🔮 前景展望AI analysis grounded in cited sources

⏳ 時間線