Updated OpenAI-compatible API guidance to prefer bedrock-runtime: Made bedrock-runtime the recommended path for new Responses API and Chat Completions API integrations. Added neutral canonical pages and redirects from the previous bedrock-mantle -named pag

Amazon Bedrock Release Notes

Amazon Bedrock now provides the OpenAI Responses API on both the bedrock-runtime and bedrock-mantle endpoints. Use bedrock-runtime for new applications and bedrock-mantle for compatibility with existing applications.

Endpoints

• bedrock-runtime: https://bedrock-runtime.{region}.amazonaws.com/openai/v1, recommended for new applications.
• bedrock-mantle: https://bedrock-mantle.{region}.api.aws/v1, for existing applications requiring Mantle-only support.

Key Benefits

• Asynchronous inference on bedrock-mantle only.
• Stateful conversation management.
• Simplified tool use.
• Flexible response modes: streaming and non-streaming.
• Easy migration from existing OpenAI SDK codebases.

Quotas

• bedrock-runtime: Tokens-per-minute and tokens-per-day quotas apply.
• bedrock-mantle: Refer to the specific quotas documentation.

Supported Regions and Endpoints

The Responses API is available on bedrock-runtime in all AWS Regions where the endpoint is available, including AWS GovCloud (US) Regions. For bedrock-mantle, see the list of supported regions and endpoints.

Prerequisites

• Authentication: Use Amazon Bedrock API key or AWS credentials.
• Environment variables: Set OPENAI_API_KEY and OPENAI_BASE_URL.
• Permissions: Required actions depend on the endpoint.

Migrating from bedrock-mantle to bedrock-runtime

• Confirm model and feature support on bedrock-runtime.
• Update Base URL, Model value, and Project settings.
• Test requests for differences in behavior.

Responses API

The Responses API provides stateful conversation management and supports streaming and multi-turn interactions. Note that not all models support the Responses API on both endpoints.

How the Responses API Stores Conversation State

Conversation state is stored for 30 days by default. Set store to false to disable data retention.

Basic Request Example

Examples provided for both bedrock-runtime and bedrock-mantle endpoints using the OpenAI SDK and HTTP requests.

Using the Responses API on the bedrock-runtime endpoint

• Requests are always synchronous.
• Server-side tool use and pre-configured tools are not available.
• Only the default project is supported.
• Guardrails do not apply to the Responses API.

Monitoring and Cost

Every request is synchronous, and usage is attributed to the inference target.

Source:

If you need further guidance on AWS, our experts are available at AWS@westloop.io. You may also reach us by submitting the Contact Us form.