Amazon Web Services has announced significant expansions to Amazon Bedrock, its managed service for accessing foundation models, adding new capabilities that strengthen its position in the enterprise AI market.
New Foundation Models Added
Amazon Bedrock now offers access to an expanded roster of leading AI models:
Anthropic Claude 2.1
The latest Claude model with 200K context window is now available through Bedrock, giving enterprises access to one of the most capable AI assistants with full AWS security integration.
Meta Llama 2
All Llama 2 model sizes (7B, 13B, 70B) are available, providing powerful open-weight alternatives for organizations with specific requirements.
Stability AI Models
Stable Diffusion XL for image generation joins the platform, enabling enterprise-grade image creation workflows.
Amazon Titan Expansion
AWS’s own Titan family grows with:
- Titan Text Express for faster, cost-effective text generation
- Titan Image Generator for enterprise image creation
- Titan Embeddings for semantic search and RAG applications
Agents for Amazon Bedrock
A major new capability allows organizations to build AI agents that can:
- Execute multi-step tasks automatically
- Access company knowledge bases
- Call APIs and take actions
- Maintain context across interactions
Agents can be configured through a simple interface without extensive coding, democratizing agent development for enterprises.
Knowledge Bases Integration
New knowledge base functionality enables:
- Automatic chunking and embedding of documents
- RAG (Retrieval Augmented Generation) pipelines
- Integration with Amazon OpenSearch and other data stores
- Reduced hallucination through grounded responses
Enhanced Security and Governance
Guardrails for Bedrock
Organizations can now implement content filtering and safety measures:
- Block sensitive topics
- Filter harmful content
- Customize for industry requirements
- Maintain audit trails
Model Evaluation
New tools for evaluating model performance:
- Automatic benchmarking
- Human evaluation workflows
- A/B testing capabilities
- Quality monitoring
Pricing and Availability
Bedrock uses a pay-per-use model with no upfront commitments:
| Model | Input (per 1K tokens) | Output (per 1K tokens) |
|---|---|---|
| Claude 2.1 | $0.008 | $0.024 |
| Titan Text | $0.0003 | $0.0004 |
| Llama 2 70B | $0.00195 | $0.00256 |
Enterprise Adoption Growing
Several major enterprises have already adopted Bedrock:
- Pfizer for drug discovery research
- BMW for customer service automation
- Nasdaq for financial document analysis
- Lonely Planet for content generation
What This Means
Amazon Bedrock’s expansion positions AWS as a serious contender in enterprise AI. The combination of multiple model providers, enterprise security, and native AWS integration provides a compelling option for organizations already invested in the AWS ecosystem.