Skip to main content
AWS Bedrock provides serverless access to foundation models from Anthropic, Meta, Amazon, Cohere, and other providers. It offers enterprise-grade security, compliance, and integration with AWS services.

​
Available Models

​
Anthropic Claude on Bedrock

​
Claude 4 Series (Latest)

  • anthropic.claude-opus-4-6-v1:0 - Most capable (200K context)
  • anthropic.claude-opus-4-1-20250805-v1:0 - Enhanced flagship (200K)
  • anthropic.claude-sonnet-4-5-20250929-v1:0 - Balanced (200K)
  • anthropic.claude-sonnet-4-20250514-v1:0 - Efficient (200K)
  • anthropic.claude-haiku-4-5-20251001-v1:0 - Fast (200K)

​
Claude 3 Series

  • anthropic.claude-3-7-sonnet-20250219-v1:0 - Enhanced (200K)
  • anthropic.claude-3-5-sonnet-20241022-v2:0 - Intelligent (200K)
  • anthropic.claude-3-5-haiku-20241022-v1:0 - Fast (200K)
  • anthropic.claude-3-opus-20240229-v1:0 - Powerful (200K)

​
Amazon Nova (Latest)

  • us.amazon.nova-premier-v1:0 - Premier multimodal (300K)
  • us.amazon.nova-pro-v1:0 - High-capability (300K)
  • us.amazon.nova-lite-v1:0 - Fast and cost-effective (300K)
  • us.amazon.nova-micro-v1:0 - Ultra-fast text-only (128K)

​
Meta Llama

  • meta.llama3-1-405b-instruct-v1:0 - Largest open model (128K)
  • meta.llama3-1-70b-instruct-v1:0 - Powerful (128K)
  • meta.llama3-2-90b-instruct-v1:0 - Multimodal (128K)
  • meta.llama3-2-11b-instruct-v1:0 - Multimodal (128K)

​
Other Providers

  • Cohere Command R/R+ - RAG-optimized (128K)
  • Mistral Large/Small - European AI (32K)
  • Amazon Titan - Amazon’s models (8K-300K)
All Claude and Nova models support tool calling and multimodal input.

​
Prerequisites

AWS Bedrock requires setting up infrastructure. This is more complex than direct API access.
Before using Bedrock with Forge:
  1. AWS Account: Active account with billing enabled
  2. Model Access: Request access to models in Bedrock console
  3. Bedrock Access Gateway: Deploy gateway infrastructure
  4. API Gateway: CloudFormation stack with API key

​
Setup Steps

​
Step 1: Deploy Bedrock Access Gateway

AWS Bedrock doesn’t provide a simple REST API. You need to deploy the Bedrock Access Gateway:
1

Clone the Repository

git clone https://github.com/aws-samples/bedrock-access-gateway.git
cd bedrock-access-gateway
2

Create API Key in Secrets Manager

  1. Go to AWS Secrets Manager
  2. Create a new secret
  3. Choose “Other type of secret”
  4. Add key-value pair: apiKey = your chosen API key
  5. Name it (e.g., bedrock-api-key)
  6. Note the ARN for next step
3

Deploy CloudFormation Stack

Follow the instructions in the gateway repository to:
  1. Package the Lambda function
  2. Deploy the CloudFormation template
  3. Pass your Secrets Manager ARN
  4. Wait for stack creation to complete
4

Get API Base URL

After deployment completes:
  1. Go to CloudFormation Console
  2. Select your stack
  3. Go to “Outputs” tab
  4. Copy the ApiUrl value (your API base URL)

​
Step 2: Request Model Access

1

Open Bedrock Console

Navigate to AWS Bedrock in your region.
2

Request Model Access

  1. Go to “Model access” in the left sidebar
  2. Click “Manage model access”
  3. Select the models you want to use:
    • Anthropic Claude models
    • Amazon Nova models
    • Meta Llama models
    • Others as needed
  4. Click “Request model access”
  5. Wait for approval (usually instant for most models)

​
Step 3: Configure Forge

1

Run Provider Login

forge provider login
2

Select OpenAI-Compatible Provider

Since Bedrock Gateway provides OpenAI-compatible API:
  1. Select “OpenAI-Compatible” from the list
  2. Enter your API base URL from Step 1
  3. Enter the API key from Secrets Manager
3

Configure Model

Set your default model in forge.yaml:
model: anthropic.claude-sonnet-4-20250514
Note: Use the full Bedrock model ID (with vendor prefix)
4

Verify Connection

forge
Try a prompt:
> Explain how AWS Bedrock works

​
Configuration

​
API Endpoint

Your Bedrock Access Gateway URL format: 
https://{api-id}.execute-api.{region}.amazonaws.com/prod

​
Model IDs

Use the full Bedrock model identifier:
  • anthropic.claude-sonnet-4-20250514-v1:0
  • us.amazon.nova-pro-v1:0
  • meta.llama3-1-70b-instruct-v1:0
List available models in Forge: 
/model

​
Model Selection

​
For Coding and Agents

Best Performance:
  • anthropic.claude-opus-4-6-v1:0 - Most capable
  • anthropic.claude-sonnet-4-5-20250929-v1:0 - Balanced
Budget-Friendly:
  • anthropic.claude-haiku-4-5-20251001-v1:0 - Fast and cheap
  • us.amazon.nova-lite-v1:0 - Cost-effective

​
For Maximum Context

  • 300K: Amazon Nova models
  • 200K: Claude 3 and 4 models
  • 128K: Llama 3.1/3.2, Cohere Command R

​
For Multimodal Tasks

  • Claude 4 - Excellent vision capabilities
  • Amazon Nova Pro/Premier - High-quality multimodal
  • Llama 3.2 90B/11B - Open-source multimodal

​
Switching Models

/model anthropic.claude-opus-4-6-v1:0

​
Regional Deployment

Bedrock is available in multiple AWS regions: US Regions:
  • us-east-1 (N. Virginia) - Most models
  • us-west-2 (Oregon) - Most models
Europe:
  • eu-west-1 (Ireland)
  • eu-central-1 (Frankfurt)
Asia Pacific:
  • ap-southeast-1 (Singapore)
  • ap-northeast-1 (Tokyo)
Model availability varies by region. Check the Bedrock documentation for details.

​
Features

​
Enterprise Security

  • VPC Endpoints: Private connectivity
  • IAM Policies: Fine-grained access control
  • CloudTrail: Full audit logging
  • KMS Encryption: Customer-managed keys

​
Compliance

  • SOC 2
  • HIPAA
  • GDPR
  • PCI DSS
  • FedRAMP (select regions)

​
AWS Integration

  • CloudWatch: Monitoring and metrics
  • S3: Data storage and retrieval
  • Lambda: Serverless integration
  • SageMaker: ML pipeline integration

​
Best Practices

​
Security

Store API keys in AWS Secrets Manager, never in code or environment variables.
Access Control:
  • Use IAM roles with least privilege
  • Rotate API keys regularly
  • Enable CloudTrail logging
  • Use VPC endpoints for private access

​
Cost Optimization

Model Selection:
  • Use Haiku/Nova Lite for simple tasks
  • Use Sonnet for balanced workloads
  • Reserve Opus/Premier for complex tasks
Infrastructure:
  • Use Lambda provisioned concurrency if needed
  • Monitor API Gateway costs
  • Set up billing alerts
Token Management:
  • Minimize context size
  • Cache common prompts
  • Use streaming for long responses

​
Rate Limits

Bedrock has:
  • Requests per minute: Varies by model and region
  • Tokens per minute: Varies by model
  • Concurrent requests: Service quotas apply
Request quota increases through AWS Support.

​
Troubleshooting

​
Gateway Deployment Fails

If CloudFormation deployment fails:
  1. Check you have sufficient IAM permissions
  2. Verify Secrets Manager ARN is correct
  3. Check CloudFormation events for specific errors
  4. Ensure Lambda execution role has Bedrock access

​
Model Access Denied

If you can’t access a model:
  1. Verify model access was granted in Bedrock console
  2. Check you’re using the correct model ID
  3. Ensure your region supports the model
  4. Wait a few minutes after granting access

​
API Key Authentication Failed

If authentication fails:
  1. Verify API key matches Secrets Manager
  2. Check API Gateway is deployed correctly
  3. Test the gateway URL directly
  4. Review CloudWatch logs for errors

​
Quota Exceeded

If you hit limits:
  1. Check current quotas in Service Quotas Console
  2. Request quota increase
  3. Implement exponential backoff
  4. Distribute load across regions

​
Region Not Supported

If model isn’t available:
  1. Check model availability by region
  2. Switch to a supported region
  3. Redeploy gateway in new region
  4. Update Forge configuration

​
Deprecated: Environment Variable Setup

Using environment variables is deprecated. Please use forge provider login instead.
For backward compatibility with the gateway: 
# .env
OPENAI_API_KEY=your-bedrock-gateway-api-key
OPENAI_URL=your-bedrock-gateway-base-url

# forge.yaml
model: anthropic.claude-sonnet-4-20250514

​
Cost Comparison

Bedrock pricing varies by model: Claude Models:
  • Similar pricing to direct Anthropic API
  • No additional Bedrock fees
  • API Gateway costs apply ($3.50/million requests)
  • Lambda costs apply (minimal for this use case)
Amazon Nova:
  • Competitive pricing
  • No API Gateway fees when using direct SDK
Meta Llama:
  • Open-source models, lower cost
  • Pay only for infrastructure
See Bedrock Pricing for details.

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us