product updateAnthropic

Claude Opus 5.5 and Sonnet 5.5 now on Amazon Bedrock in AWS GovCloud (US), with Claude Code support

TL;DR

Claude Opus 5.5 and Claude Sonnet 5.5 are available on Amazon Bedrock in AWS GovCloud (US) Regions. AWS published a setup guide for running Anthropic's Claude Code against them for regulated workloads, including ITAR. Pricing, context window and benchmark figures were not disclosed.

3 min read
0

Anthropic's Claude Opus 5.5 and Claude Sonnet 5.5 are now available on Amazon Bedrock in the AWS GovCloud (US) Regions, according to an AWS Machine Learning blog post. The post walks through running Claude Code, Anthropic's agentic coding tool, against those models for workloads with regulatory requirements, including International Traffic in Arms Regulations (ITAR).

Compliance status

AWS lists the following for models on Bedrock in GovCloud (US):

  • Claude Sonnet 5: FedRAMP Class D (formerly High) certification and DoD Impact Level 4 and 5 (IL4/IL5) authorization.
  • Claude Opus 5.5 and Claude Sonnet 5.5: FedRAMP Class D certification. The post does not state IL4/IL5 authorization for these two models.

AWS tells readers to verify current model-level certification on the Amazon Bedrock models compliance page. It also says the approach may not suit all organizations or compliance programs and should be checked against each organization's own obligations.

Two endpoints, one inference engine

Bedrock in GovCloud (US) exposes two endpoints, both powered by the same Mantle inference engine with a Zero Operator Access (ZOA) architecture:

Endpoint API Regions Notable features
bedrock-runtime AWS SDK (InvokeModel, Converse) GovCloud US-West and US-East Guardrails, Knowledge Bases, Agents, invocation logging
bedrock-mantle Anthropic Messages API (native) GovCloud US-West only Server-side tools, background inference, Projects

Both endpoints serve Claude Opus 5.5, Claude Sonnet 5.5 and Claude Sonnet 5. AWS recommends bedrock-runtime for most new applications, particularly those that need audit trails. Guardrails and invocation logging are available only through bedrock-runtime. AWS also says Bedrock does not store, log or use customer content to train AWS models, and does not share it with third parties.

Claude Code setup

Claude Code runs in the terminal, in IDEs including VS Code and JetBrains, and in the background through the Claude Agent SDK. AWS describes three configuration paths:

  1. Interactive wizard: run /login, select 3rd-party platform, then Amazon Bedrock, choose an authentication method, set region us-gov-west-1 and pin models. /setup-bedrock reopens the wizard.
  2. Environment variables: set CLAUDE_CODE_USE_BEDROCK=1 and AWS_REGION=us-gov-west-1. The model IDs are us-gov.anthropic.claude-sonnet-5-5 and us-gov.anthropic.claude-opus-5-5. Team-wide pinning uses ANTHROPIC_DEFAULT_OPUS_MODEL and ANTHROPIC_DEFAULT_SONNET_MODEL.
  3. Mantle endpoint: set CLAUDE_CODE_USE_MANTLE=1. Setting both Mantle and Bedrock flags enables both endpoints in one session.

For bedrock-runtime, the minimum IAM permissions are bedrock:InvokeModel, bedrock:InvokeModelWithResponseStream, bedrock:ListInferenceProfiles and bedrock:GetInferenceProfile. For Mantle, they are bedrock-mantle:CreateInference, GetProject, ListProjects and ListModels. The AmazonBedrockMantleInferenceAccess managed policy is an alternative. Model access for all three Claude models must be enabled in the GovCloud account.

AWS lists these Claude Code capabilities: multi-file edits, test and lint execution, Git history search, merge conflict resolution, pull request creation, sub-agents, MCP connections (AWS CLI, Terraform, Kubernetes), CLAUDE.md memory files, skills, hooks, and CI/CD integration through GitHub Actions or GitLab CI/CD.

Not disclosed

The post does not give GovCloud pricing per 1M tokens, context window sizes, benchmark scores or a launch date for the GovCloud availability.

What this means

The main change is that agentic coding tooling now has a documented path into FedRAMP Class D environments, with ITAR workloads named as a target. Regulated teams have mostly been limited to chat-style or API-only use of frontier models. The gap that matters is certification: only Sonnet 5 is listed with IL4/IL5 authorization, so DoD teams needing those levels should confirm status for the 5.5 models before adopting them.

The endpoint split also forces a trade-off. Teams that need audit trails, Guardrails or invocation logging must use bedrock-runtime. Server-side tools, background inference and Projects are available only through Mantle, and only in US-West. Architects of compliance-sensitive deployments will have to choose between governance controls and newer agent features until the two surfaces converge.

Related Articles

research

Anthropic mathematician Levent Alpöge claims Claude-assisted proof of the Riemann hypothesis, posts it in Lean

Levent Alpöge, a mathematician employed by Anthropic, says he has proven the Riemann hypothesis, first proposed in 1859, working with Claude. He posted the proof coded in Lean, and experts have yet to confirm that it is correct.

product update

Kimi Code CLI v2.7.0 adds native Amazon Bedrock provider with standard AWS credentials

Kimi Code CLI v2.7.0, Moonshot AI's open source coding agent, adds a native bedrock provider. It resolves the standard AWS credential chain, including named profiles and SSO, so users no longer need a separate Amazon Bedrock API key. AWS demonstrates the setup with Kimi K3.1 Pro Highspeed.

product update

Zed v1.29.1 adds Claude Fable 5.2 for Anthropic BYOK users and a Git Panel branch mode

Zed v1.29.1 adds support for Claude Fable 5.2 for Anthropic BYOK users, plus Claude Mythos 5.2 for organizations with access. The release also adds a Git Panel branch mode, a bookmark picker, configurable Git date formats and a setting to turn off automatic agent thread titles.

product update

Cline v4.1.27 fixes reasoning-effort errors that caused GPT-6 and gpt-5.6 tool-call rejections

Cline v4.1.27 fixes several reasoning-effort bugs that caused GPT-6 and gpt-5.6 endpoints to reject tool calls. It also refreshes the model catalog with 94 new providers and changes default models for 67 others.

Comments

Loading...