What's New

AWS has made Claude Sonnet 5 available on its platform as of June 30, 2026, marking the first Sonnet-tier model from Anthropic's latest generation. It delivers near-Opus-level intelligence at Sonnet pricing, targeting coding, agentic workflows, and professional knowledge work at scale. Customers can access it through two distinct paths: Amazon Bedrock for AWS-native, data-residency-compliant deployments, or Claude Platform on AWS for a direct Anthropic platform experience unified with AWS billing.

How It Works

• Model architecture: Claude Sonnet 5 features a 1M-token context window, up to 128K output tokens, and always-on adaptive reasoning (configurable effort level), with a knowledge cutoff of January 2026.
• Amazon Bedrock access: The model is served via two endpoints — bedrock-runtime (model ID: anthropic.claude-sonnet-5) and the newer bedrock-mantle endpoint — supporting in-region, geo-routed, and global cross-region inference options.
• Bedrock-native features: When accessed through bedrock-runtime, the model integrates with Guardrails, Knowledge Bases, Agents, Flows, prompt caching (min 4,096 tokens, up to 4 checkpoints, TTL of 5 min or 1 hour), structured outputs, and computer use.
• Claude Platform on AWS: Signing up via the AWS Console provisions a new Anthropic organization linked to your AWS account, enabling access to Anthropic's native APIs, Claude Console, and workspace management — with billing and authentication unified through AWS.
• Data handling distinction: Bedrock keeps all data within AWS infrastructure; Claude Platform on AWS routes content through Anthropic's systems outside of AWS, governed by Anthropic's data use policies.
• Prompt caching: Supported on both endpoints, with cache checkpoints applicable to system, messages, and tools fields, enabling faster and cheaper repeated inference over long contexts.
• Computer use: Supported via both bedrock-runtime and bedrock-mantle endpoints using the computer-use-2025-11-24 tool type.

Why It's Important

• Capability uplift at existing price tier: Claude Sonnet 5 brings near-Opus intelligence to the Sonnet pricing tier, meaning teams can access significantly more capable reasoning and coding without moving to a more expensive model class.
• Agentic reliability: Improved tool-calling precision, multi-step state retention, and autonomous error recovery mean agentic pipelines complete more runs successfully with less human intervention — directly reducing operational overhead.
• Large codebase handling: The ability to navigate large codebases, execute multi-file changes, and complete debugging end-to-end reduces iteration cycles in software development workflows.
• Enterprise data compliance: Bedrock's regional data residency and geo-routing options make it viable for regulated industries that cannot allow data to leave specific geographic boundaries.
• Unified billing simplicity: Claude Platform on AWS removes the need for a separate Anthropic billing relationship, consolidating AI spend within existing AWS accounts and procurement workflows.
• 1M token context: The massive context window enables processing of entire codebases, lengthy legal documents, or large datasets in a single inference call, unlocking use cases previously requiring chunking pipelines.

How It's Different

• First next-gen Sonnet: Claude Sonnet 5 is the first Sonnet model from Anthropic's latest model generation, representing a generational leap rather than an incremental update within the same family.
• Always-on adaptive reasoning: Unlike prior Sonnet models, reasoning is always active and cannot be disabled — only the effort level is configurable — making it inherently more capable on complex tasks without manual prompting strategies.
• Dual access paths: No other Anthropic model on AWS currently offers both a fully AWS-native Bedrock path and a direct Anthropic-platform path (Claude Platform on AWS) from the same AWS Console, giving teams flexibility in how they integrate.
• bedrock-mantle endpoint: Claude Sonnet 5 is accessible via the newer bedrock-mantle endpoint (in addition to bedrock-runtime), which exposes an Anthropic-native Messages API surface directly within AWS infrastructure.
• Near-Opus intelligence at Sonnet cost: Prior Sonnet models were positioned as mid-tier; Sonnet 5 explicitly targets near-Opus performance, collapsing the capability gap between tiers without the associated cost increase.
• 128K output tokens: The maximum output length is substantially higher than earlier Sonnet models, enabling generation of long-form documents, full code files, and detailed structured analyses in a single response.

When to Prefer It

• Large-scale software development: When your team needs a model that can autonomously navigate multi-file codebases, execute refactoring across many files, and debug with minimal back-and-forth, Sonnet 5 is the right choice over lighter models.
• Production agentic pipelines: When building agents that must call tools reliably, maintain state across many steps, and self-correct on errors — especially where failed runs are costly — Sonnet 5's improved agentic reliability justifies its use over prior Sonnet versions.
• Regulated or data-residency-constrained workloads: When your organization requires data to remain within a specific AWS region or geography (e.g., EU data sovereignty), use the Amazon Bedrock path with in-region or geo-routing inference.
• Teams already using Anthropic's platform: When developers want Anthropic's native console, APIs, and features but need AWS-unified billing and authentication, Claude Platform on AWS eliminates the need to manage a separate vendor relationship.
• Long-context knowledge work: When tasks involve processing large documents, lengthy transcripts, or full codebases in a single pass — such as contract analysis, audit log review, or codebase summarization — the 1M token context window makes Sonnet 5 the practical choice.
• Cost-sensitive high-capability use cases: When Opus-tier pricing is prohibitive but you need near-Opus reasoning quality for professional tasks like structured analysis, report generation, or complex Q&A, Sonnet 5 offers the best capability-per-dollar trade-off.
• Guardrails and compliance-integrated deployments: When enterprise governance requires content filtering, prompt safety controls, or integration with internal knowledge bases, the Bedrock path with native Guardrails and Knowledge Base support is preferred.

Availability

• GA status: Generally available as of June 25, 2026 (model launch date), with the AWS announcement published June 30, 2026.
• Regional availability: Available via Amazon Bedrock with in-region, geo-routed (US, EU, Japan, Australia), and global cross-region inference; specific supported regions are listed in the Bedrock regional availability documentation.
• GovCloud: Accessible in AWS GovCloud (US) after accepting the EULA in a linked standard account (us-east-1 or us-west-2) and enabling the model via the GovCloud Model Access page.
• Model IDs: anthropic.claude-sonnet-5 on both bedrock-runtime and bedrock-mantle endpoints; geo inference ID us.anthropic.claude-sonnet-5; global inference ID global.anthropic.claude-sonnet-5.
• Pricing: Follows standard Amazon Bedrock pricing for Sonnet-tier models; refer to the Amazon Bedrock Pricing page for per-token rates. Claude Platform on AWS billing is unified through AWS Marketplace.
• Context and output limits: 1M token context window; 128K maximum output tokens per request.
• Key limitation — Claude Platform on AWS: Content processed via Claude Platform on AWS is handled by Anthropic outside of AWS infrastructure; teams with strict data residency requirements should use the Amazon Bedrock path instead.
• Model lifecycle: Active, with no announced end-of-life date.

Related Resources

• https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-sonnet-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