Skip to main content
Bedrock Chat supports multiple foundation models from Amazon Bedrock. Configure which models are available to users and set default model preferences.

​
Enable Bedrock Model Access

Before deploying Bedrock Chat, enable model access in Amazon Bedrock:
  1. Navigate to the Bedrock Model Access page in us-east-1
  2. Click Manage model access
  3. Select all models you wish to use
  4. Click Save changes
Without enabling model access in Bedrock, the models will not be available in Bedrock Chat even if configured.

​
Supported Models

Bedrock Chat supports the following foundation models:

​
Claude Models (Anthropic)

  • claude-v4-opus
  • claude-v4.1-opus
  • claude-v4.5-opus
  • claude-v4-sonnet
  • claude-v4.5-sonnet
  • claude-v4.5-haiku
  • claude-v3-opus
  • claude-v3.5-sonnet
  • claude-v3.5-sonnet-v2
  • claude-v3.7-sonnet (supports extended thinking up to 64k tokens)
  • claude-v3-haiku
  • claude-v3.5-haiku

​
Amazon Nova Models

  • amazon-nova-pro
  • amazon-nova-lite
  • amazon-nova-micro

​
Mistral Models

  • mistral-7b-instruct
  • mixtral-8x7b-instruct
  • mistral-large
  • mistral-large-2

​
DeepSeek Models

  • deepseek-r1

​
Meta Llama Models

  • llama3-3-70b-instruct
  • llama3-2-1b-instruct
  • llama3-2-3b-instruct
  • llama3-2-11b-instruct
  • llama3-2-90b-instruct

​
Other Models

  • gpt-oss-20b
  • gpt-oss-120b

​
Global Model Configuration

Control which models appear in the UI for all users.
​
globalAvailableModels
array
default:"[]"
List of model IDs to enable globally. Empty array enables all models. When set, only specified models appear in dropdown menus.
Configure in cdk.json - Enable all models (default): 
{
  "context": {
    "globalAvailableModels": []
  }
}
Configure in cdk.json - Restrict to specific models: 
{
  "context": {
    "globalAvailableModels": [
      "claude-v3.7-sonnet",
      "claude-v3.5-sonnet",
      "amazon-nova-pro",
      "amazon-nova-lite",
      "llama3-3-70b-instruct"
    ]
  }
}
Configure in parameter.ts: 
bedrockChatParams.set("default", {
  globalAvailableModels: [
    "claude-v3.7-sonnet",
    "claude-v3.5-sonnet",
    "amazon-nova-pro",
    "amazon-nova-lite",
    "llama3-3-70b-instruct",
  ],
});
Deployment script: 
./bin.sh --cdk-json-override '{
  "context": {
    "globalAvailableModels": [
      "claude-v3.7-sonnet",
      "claude-v3.5-sonnet",
      "amazon-nova-pro"
    ]
  }
}'
Only models enabled in the Bedrock console AND included in globalAvailableModels will be available to users.

​
Default Model Selection

Specify which model is pre-selected when users start a new chat.
​
defaultModel
string
default:"claude-v3.7-sonnet"
Model ID pre-selected in the chat UI when users start a new conversation.
Configure in parameter.ts: 
bedrockChatParams.set("default", {
  defaultModel: "amazon-nova-pro",
});
Ensure the defaultModel is enabled in Bedrock console and included in globalAvailableModels (if specified). Otherwise, users may see an error when starting conversations.

​
Title Generation Model

Specify which model automatically generates conversation titles.
​
titleModel
string
default:"claude-v3-haiku"
Model used to generate short conversation titles. Defaults to defaultModel if not specified, then falls back to claude-v3-haiku.
Configure in parameter.ts: 
bedrockChatParams.set("default", {
  titleModel: "amazon-nova-lite", // Cost-optimized for title generation
});
Use a smaller, cost-effective model like claude-v3-haiku, amazon-nova-lite, or amazon-nova-micro for title generation to reduce costs without impacting user experience.
Example configuration: 
bedrockChatParams.set("default", {
  defaultModel: "claude-v3.7-sonnet",     // Premium model for chat
  titleModel: "amazon-nova-lite",          // Budget model for titles
  globalAvailableModels: [
    "claude-v3.7-sonnet",
    "claude-v3.5-sonnet",
    "amazon-nova-pro",
    "amazon-nova-lite",
  ],
});

​
Bedrock Region Configuration

Specify which AWS region to use for Amazon Bedrock API calls.
​
bedrockRegion
string
default:"us-east-1"
AWS region where Amazon Bedrock APIs are available. Must be a region where Bedrock is supported.
Configure in cdk.json: 
{
  "context": {
    "bedrockRegion": "us-west-2"
  }
}
Configure in parameter.ts: 
bedrockChatParams.set("default", {
  bedrockRegion: "us-west-2",
});
Deployment script: 
./bin.sh --bedrock-region "us-west-2"
Amazon Bedrock is not available in all AWS regions. Check the Bedrock documentation for supported regions.

​
Cross-Region and Global Inference

Amazon Bedrock can dynamically route inference requests across regions for improved throughput and resilience.

​
Global Inference

​
enableBedrockGlobalInference
boolean
default:"true"
Enable global inference to route requests to the optimal region worldwide based on latency and availability.
Configure in cdk.json: 
{
  "context": {
    "enableBedrockGlobalInference": true
  }
}

​
Cross-Region Inference

​
enableBedrockCrossRegionInference
boolean
default:"true"
Enable cross-region inference to route requests within the same geographic area (e.g., within US regions).
Configure in cdk.json: 
{
  "context": {
    "enableBedrockCrossRegionInference": true
  }
}
Configure both in parameter.ts: 
bedrockChatParams.set("default", {
  bedrockRegion: "us-east-1",
  enableBedrockGlobalInference: true,
  enableBedrockCrossRegionInference: true,
});
Some AWS Service Control Policies (SCPs) may restrict global or cross-region inference. Configure these settings based on your organization’s policies.

​
Model Parameters

Users can configure generation parameters for each conversation:

​
Temperature

  • Range: 0 to 1
  • Step: 0.05
  • Default: 0.6
  • Purpose: Controls randomness in responses. Lower values make output more deterministic.

​
Max Tokens

  • Range: 1 to 64,000
  • Step: 10
  • Default: 2,000
  • Purpose: Maximum number of tokens to generate in the response.
Claude 3.7 with extended thinking supports up to 64,000 tokens output.

​
Top P

  • Range: 0 to 1
  • Step: 0.001
  • Default: 0.999
  • Purpose: Nucleus sampling parameter controlling diversity.

​
Top K

  • Range: 0 to 500
  • Step: 1
  • Default: 128
  • Purpose: Limits the number of highest probability tokens to consider.

​
Reasoning Budget Tokens

  • Range: 1,024 to 64,000
  • Step: 10
  • Default: 1,024
  • Purpose: Token budget allocated for reasoning and thinking (Claude extended thinking).

​
Multi-Environment Model Configuration

Configure different models for different environments: 
// Development - cost-optimized
bedrockChatParams.set("dev", {
  bedrockRegion: "us-east-1",
  defaultModel: "amazon-nova-lite",
  titleModel: "amazon-nova-micro",
  globalAvailableModels: [
    "amazon-nova-lite",
    "amazon-nova-micro",
    "claude-v3-haiku",
  ],
});

// Production - full model access
bedrockChatParams.set("prod", {
  bedrockRegion: "us-west-2",
  defaultModel: "claude-v3.7-sonnet",
  titleModel: "claude-v3-haiku",
  enableBedrockGlobalInference: true,
  enableBedrockCrossRegionInference: true,
  globalAvailableModels: [
    "claude-v3.7-sonnet",
    "claude-v3.5-sonnet",
    "claude-v3-opus",
    "amazon-nova-pro",
    "amazon-nova-lite",
    "llama3-3-70b-instruct",
  ],
});
Deploy specific environment: 
npx cdk deploy --all -c envName=prod

​
Model Selection Best Practices

  1. Cost optimization: Use smaller models (amazon-nova-lite, claude-v3-haiku) for title generation
  2. Performance: Use latest Claude models for best reasoning and accuracy
  3. Compliance: Restrict globalAvailableModels to organization-approved models only
  4. Regional availability: Choose bedrockRegion with best model availability for your use case
  5. Throughput: Enable global/cross-region inference for high-traffic deployments

Build docs developers (and LLMs) love

Get started for freeTalk to us