What's New in Amazon Bedrock: 100+ Models, Smarter Inference, and Enterprise Data Integration

Amazon Web Services is making significant moves to democratize generative AI adoption. At its annual re:Invent conference, AWS announced a major update to Amazon Bedrock—its managed foundation model platform—introducing over 100 new accessible models, advanced inference optimization, and powerful data processing capabilities designed to help enterprises accelerate AI deployment.

The Model Explosion: Access to 100+ Foundation Models

The breadth of available models is expanding dramatically. Beyond AWS’s own Amazon Nova foundation models, the ecosystem now includes contributions from leading AI labs. Luma AI’s Ray 2 model brings video generation capabilities to production environments, enabling users to create high-quality video content from text and images with realistic physics and consistent character behavior. This opens possibilities for marketing teams, architects, and designers looking to prototype visual concepts rapidly.

Code-focused teams now have access to poolside’s malibu and point models through Amazon Bedrock, tools specifically designed for software engineering tasks like code generation, testing, and documentation. These models can be fine-tuned on enterprise codebases, allowing companies to build AI assistants adapted to their specific development practices and standards.

Stability AI’s Stable Diffusion 3.5 Large joins the platform for image generation workflows. The model supports diverse artistic styles and accelerates concept art creation for industries from gaming to retail.

Beyond these flagship additions, the Amazon Bedrock Marketplace now catalogs over 100 models—including specialized options for finance (Writer’s Palmyra-Fin), translation (Upstage’s Solar Pro), and biological research (EvolutionaryScale’s ESM3). Customers select the model matching their use case, configure infrastructure through AWS, and deploy through unified APIs with built-in governance and security.

Smarter Inference: Prompt Caching and Dynamic Routing

As models scale into production, inference costs and latency become critical constraints. Two capabilities address this directly.

Prompt Caching allows frequently reused content to be cached securely, reducing processing overhead. Early results show meaningful improvements: Adobe’s Acrobat AI Assistant experienced a 72% reduction in response time when caching prompts on Amazon Bedrock. Cost reductions can reach 90% for supported models, while latency drops by up to 85%.

Intelligent Prompt Routing handles request complexity dynamically. The system analyzes incoming prompts using advanced matching techniques and routes each to the optimal model within a family. Simple queries go to smaller, cheaper models; complex questions route to larger models. The outcome: up to 30% cost reduction while maintaining response quality. Argo Labs, a voice AI company, uses this approach to handle restaurant customer queries—directing simple yes-no questions to lightweight models while reserving compute for nuanced menu and availability inquiries.

Data Leverage: Structured Queries and Knowledge Graphs

Amazon Bedrock Knowledge Bases now supports structured data retrieval directly. Instead of converting enterprise databases into unstructured text, customers can query structured data using natural language, with the system translating queries into SQL executed against data warehouses and data lakes. Octus, a credit intelligence platform, plans to use this to let end users explore structured credit data conversationally, transforming months of integration work into days of configuration.

Knowledge graph capabilities (GraphRAG) enable enterprises to model relationships within their data automatically. BMW Group plans to implement this for its internal data assistant (MAIA), using graph databases to maintain contextual relationships between data assets and continuously improve response relevance based on actual usage patterns.

Automated Data Pipeline: From Unstructured to Structured

A new Amazon Bedrock Data Automation service transforms documents, images, audio, and video into structured formats—automatically. Banks processing loan documents, insurance companies analyzing claims, and digital asset teams managing content repositories can now extract, normalize, and structure data at scale without manual effort.

The automation includes built-in confidence scoring and grounds outputs in source material to reduce hallucination risks. Symbeo uses this for accounts payable automation—extracting data from insurance claims and medical bills faster. Tenovos uses it for semantic search, reporting 50%+ increases in content reuse.

Adoption Momentum

The installed base reflects the strategy’s success. Amazon Bedrock now serves tens of thousands of customers—growing 4.7x year-over-year. Adobe, BMW Group, Zendesk, Argo Labs, and others are already adopting these new capabilities, indicating confidence in the platform’s maturity and direction.

Availability and Rollout

Amazon Bedrock Marketplace is available immediately. Prompt caching, Intelligent Prompt Routing, Knowledge Bases enhancements (structured data and GraphRAG), and Data Automation are in preview. Luma AI, poolside, and Stability AI models are coming soon.

The bedrock news reflects AWS’s broader strategy: reduce friction for enterprises building AI applications by handling infrastructure, model selection, and cost optimization automatically. For development teams, this translates to faster prototyping, lower experimentation costs, and easier transition from proof-of-concept to production deployment.