Amazon Bedrock Adds Claude 4 and Llama 4 Support

Amazon Web Services has officially expanded its Amazon Bedrock managed AI service with native support for Anthropic's Claude 4 model family and Meta's Llama 4 series. The move positions AWS as one of the first major cloud providers to offer both cutting-edge model families through a single unified API, giving enterprise developers immediate access to the latest generation of frontier and open-weight large language models.

This dual integration signals a significant strategic shift for AWS, which has historically lagged behind Microsoft Azure and Google Cloud in the generative AI platform wars. By onboarding Claude 4 and Llama 4 simultaneously, Amazon is making an aggressive bid to become the default infrastructure layer for enterprise AI deployment.

Key Facts at a Glance

• Claude 4 family — including Claude Opus 4 and Claude Sonnet 4 — is now available natively through Amazon Bedrock's API
• Llama 4 Scout and Llama 4 Maverick, Meta's latest open-weight models, join the Bedrock model catalog
• Developers can access both model families through Bedrock's unified API without managing separate infrastructure
• Integration includes support for Bedrock Guardrails, Knowledge Bases, and Agents features
• Pricing follows Bedrock's standard on-demand and provisioned throughput billing models
• Available across multiple AWS regions including US East, US West, and EU (Frankfurt)

Claude 4 Brings Agentic AI Capabilities to Bedrock

Anthropic's Claude 4 models represent a generational leap over the Claude 3.5 series that previously dominated Bedrock's model offerings. Claude Opus 4, the flagship model, introduces what Anthropic calls 'sustained agentic performance' — the ability to work autonomously on complex, multi-step tasks for extended periods without losing coherence or accuracy.

Claude Sonnet 4, positioned as the mid-tier option, delivers what many developers consider the sweet spot between capability and cost. It outperforms Claude 3.5 Sonnet on coding benchmarks like SWE-bench by a significant margin while maintaining competitive pricing.

For Bedrock customers, the integration means these models work seamlessly with existing Bedrock Agents workflows. Enterprises already using Bedrock for retrieval-augmented generation (RAG) pipelines or automated customer service can upgrade to Claude 4 models with minimal code changes — often just swapping a model identifier in their API calls.

Meta's Llama 4 Expands Open-Weight Options

Meta's Llama 4 series brings a fundamentally different architecture to the Bedrock platform. Both Llama 4 Scout and Llama 4 Maverick use a mixture-of-experts (MoE) architecture, which activates only a subset of model parameters for each query. This design delivers stronger performance while keeping inference costs manageable.

Llama 4 Scout features 17 billion active parameters drawn from a total of 109 billion, supporting a massive 10 million token context window. Llama 4 Maverick scales up to 17 billion active parameters from 400 billion total, making it competitive with much larger dense models.

The open-weight nature of Llama 4 models offers enterprises a distinct advantage on Bedrock:

• Cost efficiency — MoE architecture means lower per-token inference costs compared to dense models of similar capability
• Customization potential — open weights enable fine-tuning through Bedrock's custom model training features
• Transparency — organizations in regulated industries can inspect model weights for compliance purposes
• Vendor flexibility — no proprietary lock-in to a single model provider's ecosystem
• Multilingual strength — Llama 4 supports over 200 languages natively, expanding global deployment options

How This Changes the Cloud AI Platform Race

The simultaneous addition of Claude 4 and Llama 4 reshapes the competitive dynamics among the 3 major cloud providers. Microsoft Azure has leaned heavily on its exclusive partnership with OpenAI, offering GPT-4o and GPT-4.1 as primary model options through Azure OpenAI Service. Google Cloud promotes its own Gemini 2.5 models through Vertex AI while also hosting third-party options.

AWS, lacking a proprietary frontier model of comparable caliber, has instead embraced a 'model marketplace' strategy. Bedrock now offers models from Anthropic, Meta, Mistral, Cohere, AI21 Labs, Stability AI, and Amazon's own Titan family. This breadth-first approach appeals to enterprises that want flexibility without managing multiple cloud relationships.

The strategy appears to be working. AWS reported that Bedrock usage grew over 3.5x year-over-year in recent earnings calls, with tens of thousands of enterprise customers actively building on the platform. Adding the 2 most anticipated model releases of 2025 should accelerate that momentum further.

Technical Integration Details for Developers

From a technical standpoint, the Claude 4 and Llama 4 integrations follow Bedrock's established patterns, minimizing the learning curve for existing users. Both model families support Bedrock's Converse API, which provides a standardized interface across all hosted models.

Key technical capabilities include:

• Streaming responses for real-time applications and chat interfaces
• Tool use and function calling natively supported across both Claude 4 and Llama 4 models
• Vision capabilities — Claude 4 and Llama 4 Maverick both accept image inputs for multimodal workflows
• Bedrock Guardrails integration for content filtering, PII redaction, and topic blocking
• Knowledge Bases compatibility for RAG-based applications with automatic chunking and embedding
• Cross-region inference for improved availability and latency optimization

Developers using the AWS SDK can switch to Claude 4 by updating their model ID parameter to reference the new model versions. For Llama 4, Bedrock handles the MoE inference routing automatically — developers interact with the same simple API regardless of the underlying architecture complexity.

Pricing and Cost Considerations

While AWS has not disclosed exact per-token pricing for all new models at the time of this writing, early indications suggest competitive positioning against rival platforms. Claude Sonnet 4 on Bedrock is expected to maintain pricing parity with Anthropic's direct API, typically around $3 per million input tokens and $15 per million output tokens.

Llama 4 models, benefiting from their open-weight status and efficient MoE architecture, should come in at significantly lower price points. Llama 4 Scout in particular targets cost-sensitive workloads where high throughput matters more than peak capability.

Bedrock's provisioned throughput option allows enterprises to reserve dedicated capacity for predictable workloads, often reducing per-token costs by 30-50% compared to on-demand pricing. This model is particularly attractive for production deployments processing millions of requests daily.

What This Means for Enterprise AI Strategy

The practical implications for businesses evaluating their AI infrastructure are substantial. Organizations no longer need to choose between the strongest proprietary model (Claude 4) and the most flexible open-weight option (Llama 4) — they can deploy both through a single platform with unified billing, security controls, and compliance frameworks.

This is especially meaningful for enterprises operating under strict data governance requirements. Bedrock processes all inference within the customer's chosen AWS region, with data never leaving that boundary. Both Claude 4 and Llama 4 on Bedrock operate under AWS's standard HIPAA, SOC 2, and ISO 27001 compliance certifications.

For development teams, the multi-model approach enables sophisticated routing strategies. A common pattern involves directing simple queries to cost-efficient Llama 4 Scout, escalating complex reasoning tasks to Claude Sonnet 4, and reserving Claude Opus 4 for the most demanding agentic workflows. Bedrock's infrastructure makes this kind of intelligent model routing straightforward to implement.

Looking Ahead: The Multi-Model Future

AWS's aggressive model onboarding strategy points toward a future where the value shifts from model access to orchestration infrastructure. As frontier models from multiple providers converge in capability, the platform that offers the best tools for managing, evaluating, and deploying across models gains a durable competitive advantage.

Amazon appears to be betting on exactly this outcome. Recent Bedrock feature additions — including model evaluation tools, automated prompt optimization, and multi-agent collaboration frameworks — all reinforce the platform's role as an orchestration layer rather than a single-model endpoint.

For the broader AI industry, the Claude 4 and Llama 4 Bedrock integration confirms that the 'model-as-a-service' era is maturing rapidly. Enterprises increasingly expect instant access to the latest models through their existing cloud relationships, with enterprise-grade security and compliance built in from day 1. The cloud providers that deliver on this expectation most effectively will capture the bulk of enterprise AI spending, projected to exceed $200 billion globally by 2028.