What's New

AWS has launched Claude Opus 5, Anthropic's most capable Opus model to date, available through both Amazon Bedrock and the new Claude Platform on AWS. The model delivers significant advances in agentic coding, long-running autonomous task execution, and deep document analysis for enterprise workloads. Notably, Amazon Bedrock enables Zero Data Retention (ZDR) by default, making Opus 5 immediately suitable for data-sensitive and regulated environments.

How It Works

• Model access via Amazon Bedrock: Claude Opus 5 is accessible using model ID anthropic.claude-opus-5 through the bedrock-runtime endpoint, with support for in-region, geographic (geo), and global inference routing options.
• Bedrock Mantle endpoint: A second endpoint (bedrock-mantle) is available at https://bedrock-mantle.{region}.api.aws/anthropic/v1/messages, supporting prompt caching, computer use, and streaming with a Messages-compatible API.
• Zero Data Retention (ZDR): On Amazon Bedrock, ZDR is enabled by default, ensuring prompts and completions are never stored by Anthropic; on Claude Platform on AWS, ZDR is available on request.
• Context and output capacity: The model supports a 1M token context window with up to 128K output tokens, enabling processing of very large documents and long-form generation in a single call.
• Adaptive reasoning: Extended thinking (adaptive reasoning) is on by default and can be disabled, though reasoning effort is capped at "high" when thinking is turned off.
• Prompt caching: Supported on the bedrock-mantle endpoint with a minimum of 512 tokens per cache checkpoint, up to 4 checkpoints per request, with TTLs of 5 minutes or 1 hour across system, messages, and tools fields.
• Claude Platform on AWS: Provides Anthropic's native API and console experience (platform.claude.com) unified with AWS billing and IAM authentication, provisioning a new Anthropic organization tied to the AWS account.
• Bedrock-native integrations: On the bedrock-runtime endpoint, Guardrails, Knowledge Bases, Agents, Flows, Prompt Management, and Model Evaluation are all supported alongside the model.

Why It's Important

• Enterprise-grade data governance out of the box: ZDR enabled by default on Bedrock removes a common procurement and compliance barrier, allowing regulated industries (finance, healthcare, legal) to adopt the model without additional negotiation.
• Long-running agentic workloads become practical: The model's ability to autonomously execute multi-hour or overnight tasks—recovering from errors and rerouting around obstacles—unlocks a new class of automation that previously required constant human supervision.
• Production-quality coding at scale: Opus 5's ability to navigate large codebases and write production-ready code with adaptive strategy makes it viable as a core component in software engineering pipelines, not just a developer assistant.
• Document-heavy enterprise work: The largest benchmark gains are on document-intensive tasks, directly addressing the needs of legal, financial, and research teams that process high volumes of complex text.
• Unified billing and authentication: Claude Platform on AWS eliminates the need for a separate Anthropic billing relationship, simplifying procurement and cost consolidation for AWS-centric organizations.

How It's Different

• ZDR by default on Bedrock: Unlike many third-party models on Bedrock where data handling requires explicit configuration, Opus 5 ships with ZDR on by default—a meaningful differentiator for compliance-first buyers.
• Dual access paths: No other Anthropic model on AWS currently offers both a fully AWS-managed Bedrock path and a native Anthropic platform path (Claude Platform on AWS) under a single AWS billing umbrella.
• 1M token context window: The context window is substantially larger than most competing frontier models available on Bedrock, enabling single-call processing of entire codebases or lengthy legal documents.
• Adaptive thinking enabled by default: Unlike models where extended reasoning must be explicitly activated, Opus 5 applies adaptive thinking automatically, optimizing reasoning depth per query without developer intervention.
• Computer use support: Opus 5 supports computer use via both bedrock-runtime and bedrock-mantle endpoints, a capability not universally available across frontier models on Bedrock.
• Geo and global inference routing: Multiple inference routing tiers (in-region, geo, global) give teams fine-grained control over the throughput-vs-residency tradeoff, which is particularly relevant for multinational deployments.

When to Prefer It

• Regulated or data-sensitive workloads: Choose Opus 5 on Amazon Bedrock when your organization requires data never to leave AWS infrastructure and ZDR compliance must be demonstrable by default.
• Autonomous, long-horizon agents: Use Opus 5 when building agents that must run for extended periods (hours or overnight), handle unexpected errors gracefully, and complete complex multi-step objectives without human intervention.
• Large codebase engineering tasks: Prefer Opus 5 when the task involves understanding, refactoring, or generating code across a large, interconnected codebase where context and strategic adaptation matter.
• Document-intensive enterprise analysis: Select Opus 5 for workflows involving large volumes of long documents—contracts, research reports, regulatory filings—where accuracy and depth of analysis are critical.
• Teams already on AWS wanting Anthropic's native experience: Use Claude Platform on AWS when developers want Anthropic's first-party API and console features but need unified AWS billing, IAM authentication, and a single vendor relationship.
• High-throughput global deployments: Use global inference routing when data residency is not a constraint and maximizing throughput and minimizing cost across all commercial regions is the priority.
• Multi-modal agentic pipelines: Choose Opus 5 when your pipeline requires processing audio, image, speech, text, and video inputs together with tool use and computer interaction in a single model call.

Availability

• GA status: Generally available as of July 24, 2026; model lifecycle is listed as Active with no announced end-of-life date.
• Knowledge cutoff: May 2026.
• Regional inference options: Supports in-region (single AWS region), geographic cross-region (US, EU, AU geographies), and global cross-region routing; specific in-region availability listed in the Bedrock regional compatibility documentation.
• Geo inference IDs: us.anthropic.claude-opus-5, eu.anthropic.claude-opus-5, au.anthropic.claude-opus-5; global ID: global.anthropic.claude-opus-5.
• GovCloud: Accessible in AWS GovCloud after accepting the EULA in a linked standard region (us-east-1 or us-west-2) and enabling via the Model Access page.
• Pricing: Refer to the Amazon Bedrock Pricing page; no flat fee announced—usage-based pricing applies (Marketplace product ID: prod-if5d653ow7ehg).
• Claude Platform on AWS ZDR: Available on request (not enabled by default, unlike Bedrock); content is processed by Anthropic outside of AWS under Anthropic's data use policies.
• Prompt caching limitation: Only available on the bedrock-mantle endpoint; minimum 512 tokens per checkpoint, maximum 4 checkpoints per request.
• Reasoning limitation: When adaptive thinking is disabled, reasoning effort is capped at "high"—full uncapped reasoning requires thinking to remain enabled.

Related Resources

• https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-opus-5.html
• https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html#model-regions-anthropic
• https://docs.aws.amazon.com/claude-platform/latest/userguide/setup.html