Claude Sonnet 5 is now available on Amazon Bedrock in AWS GovCloud (US)
Government workloads can now access Claude Sonnet 5's coding and agentic power inside GovCloud's strict data-residency boundary.
View original announcement →Visual Summary
graph TD
A{{Claude Sonnet 5 on GovCloud}}:::announced
B(Amazon Bedrock Runtime):::compute
C(Bedrock Mantle Engine):::compute
D([Guardrails]):::feature
E([Knowledge Bases]):::feature
F([Prompt Caching]):::feature
G([Data Residency]):::feature
H((Gov Agency Users)):::external
I(AWS GovCloud Infra):::storage
H ==>|"invokes"| A
A -->|"bedrock-runtime"| B
A -->|"bedrock-mantle"| C
B -->|"applies"| D
B -->|"retrieves"| E
C -->|"enables"| F
A -.->|"enforces"| G
B -->|"stays within"| I
C -->|"stays within"| I
classDef announced fill:#ff9900,stroke:#ec7211,color:#fff,font-weight:bold
classDef compute fill:#e3f2fd,stroke:#1565c0,color:#1565c0
classDef storage fill:#e8f5e9,stroke:#2e7d32,color:#2e7d32
classDef feature fill:#fff3e0,stroke:#e65100,color:#e65100
classDef external fill:#f5f5f5,stroke:#616161,color:#616161
What's New
Claude Sonnet 5, Anthropic's most capable Sonnet model to date, is now generally available on Amazon Bedrock within AWS GovCloud (US) regions, joining Claude Opus 4.8 which became available on June 30, 2026. The model is accessible via both the standard bedrock-runtime endpoints (US-West and US-East) and the next-generation bedrock-mantle endpoints (US-West), supporting the Anthropic Messages API. This expansion brings near-Opus-level intelligence to government and regulated-industry workloads requiring strict data residency and compliance controls.
How It Works
- Inference endpoints: Claude Sonnet 5 is served through
bedrock-runtimeendpoints in both GovCloud US-West and US-East, and throughbedrock-mantleendpoints in GovCloud US-West; the model ID isanthropic.claude-sonnet-5on both. - Bedrock Mantle: Amazon Bedrock's next-generation inference engine natively supports the Anthropic Messages API, enabling prompt caching (min 4,096 tokens per checkpoint, up to 4 checkpoints, with 5-minute or 1-hour TTL) and response streaming.
- Context and output: The model supports a 1M-token context window with up to 128K output tokens, and adaptive reasoning (always-on, configurable effort level) is built in.
- GovCloud model enablement: Access requires accepting the model EULA in a linked standard AWS account (us-east-1 or us-west-2) via the Bedrock console or AWS CLI, then enabling the model in the GovCloud account through the Model Access page.
- AWS-managed features: When using the bedrock-runtime endpoint, the model integrates with Guardrails, Knowledge Bases, Agents, Flows, prompt management, and structured outputs; bedrock-mantle supports a narrower feature set (streaming, abuse detection, prompt caching, count tokens, computer use).
- Data residency: All inference requests remain within AWS GovCloud infrastructure; no data crosses into commercial regions, satisfying strict regulatory requirements.
Why It's Important
- Regulated-industry access to frontier AI: Government agencies, defense contractors, and FedRAMP/ITAR-scoped organizations can now run a state-of-the-art coding and agentic model without compromising compliance posture.
- Agentic reliability at scale: Claude Sonnet 5's improved tool-calling precision and error recovery means autonomous multi-step workflows—common in government IT automation—complete successfully with fewer human interventions.
- Sovereign data processing: Regional data residency guarantees that sensitive government data, PII, and classified-adjacent workloads never leave the GovCloud boundary during inference.
- Cost-performance balance for high-volume use: Sonnet-tier pricing delivers near-Opus capability at lower cost, making it practical to run large-scale document analysis, code review pipelines, and knowledge-work automation within agency budgets.
- Unified platform integration: Access to Guardrails, Knowledge Bases, and Agents within the same GovCloud service boundary reduces architectural complexity for teams building compliant AI applications.
How It's Different
- GovCloud-native vs. commercial deployment: Unlike standard Bedrock regions, GovCloud deployments enforce strict data sovereignty, making this the only path for workloads subject to FedRAMP High, ITAR, or DoD compliance frameworks.
- Sonnet 5 vs. Opus 4.8 in GovCloud: Sonnet 5 offers a better cost-to-capability ratio for high-frequency tasks, while Opus 4.8 (also now available) targets deeper reasoning and longer autonomous runs where cost is secondary to quality.
- Bedrock Mantle vs. bedrock-runtime: Mantle's next-generation engine adds prompt caching and native Anthropic Messages API support, reducing latency and cost for repetitive or long-context workloads compared to the classic runtime.
- 1M token context vs. prior Sonnet models: The dramatically expanded context window allows entire large codebases or lengthy regulatory documents to be processed in a single call, eliminating chunking complexity.
- Adaptive reasoning always-on: Unlike some models where reasoning is a discrete mode, Sonnet 5's adaptive thinking is continuously active with a configurable effort level, providing consistent reasoning quality without explicit prompt engineering.
When to Prefer It
- Federal and government workloads: Choose Claude Sonnet 5 on GovCloud when your application must meet FedRAMP, ITAR, DoD IL, or similar compliance requirements that prohibit data leaving GovCloud boundaries.
- High-volume agentic pipelines: Prefer Sonnet 5 over Opus 4.8 when running large numbers of agentic tasks (code generation, document processing, tool-calling workflows) where cost efficiency matters and near-Opus quality is sufficient.
- Large codebase automation: Use when tasks involve navigating, refactoring, or making multi-file changes across enterprise-scale repositories, leveraging the 1M-token context to hold the full codebase in view.
- Structured knowledge work at scale: Ideal for converting unstructured government documents, reports, or data into structured formats (spreadsheets, databases, summaries) with minimal human review cycles.
- Prompt-caching-heavy applications: Select the bedrock-mantle endpoint when your application repeatedly sends large system prompts or static context (e.g., policy documents, API schemas), as prompt caching significantly reduces latency and token costs.
- Multi-step autonomous tasks requiring error recovery: Choose when building workflows that must run to completion without human intervention, such as automated compliance checks or IT remediation pipelines, where Sonnet 5's self-correction capabilities reduce failure rates.
Availability
- Status: Generally Available (GA) as of July 23, 2026.
- GovCloud regions (bedrock-runtime): AWS GovCloud US-West (
us-gov-west-1) and AWS GovCloud US-East (us-gov-east-1). - GovCloud regions (bedrock-mantle): AWS GovCloud US-West (
us-gov-west-1) only. - Co-available model: Claude Opus 4.8 is also available on the same endpoints in GovCloud, having launched June 30, 2026.
- Pricing: Refer to the Amazon Bedrock Pricing page; Sonnet-tier pricing applies (lower than Opus 4.8).
- Enablement prerequisite: EULA must be accepted in a linked standard AWS account (us-east-1 or us-west-2) before the model can be enabled in the GovCloud account; entitlement propagation may take a few minutes.
- Feature limitations on bedrock-mantle: Guardrails, Knowledge Bases, Agents, Flows, prompt management, and intelligent prompt routing are not supported on the mantle endpoint; use bedrock-runtime for full feature access.
- Model lifecycle: Active; no end-of-life date announced.
