Amazon Bedrock provides an OpenAI-compatible implementation of the Responses API for supported OpenAI models. This implementation is provided and operated by AWS, and is separate from OpenAI’s hosted Responses API.
The goal is compatibility with the OpenAI Responses API, but feature availability, behavior, model support, and release timing may differ. Availability can vary by model, AWS Region, account configuration, and AWS release status.
How it works
Requests are sent to Amazon Bedrock and authenticated using Bedrock-supported authentication methods. The OpenAI-hosted Responses API is not in the request path.
Customers manage access through their AWS account, including IAM permissions, model access, quotas, billing, regional availability, and applicable Bedrock controls.
Supported capabilities
Amazon Bedrock supports a subset of OpenAI Responses API capabilities for supported OpenAI models.
Capabilities currently supported or expected in the Bedrock implementation include:
OpenAI-compatible endpoint shape
Model catalog
Response schema
Error taxonomy
Stateful continuation
Stateless continuation
Background mode
Structured Outputs
JSON mode
Reasoning effort
Encrypted reasoning
Verbosity
Service tier
Function calling
Parallel tool calls
Text input and output
Chat Completions compatibility
Authentication
Billing
Data controls
Safety
Latency, throughput, and reliability behavior
Support path through AWS/Bedrock
Capabilities not currently available or limited
The following OpenAI-hosted API capabilities may be unavailable, limited, or behave differently in the Amazon Bedrock implementation:
Usage accounting
Stream resume
Webhooks
Compaction
Responses streaming
Realtime WebSocket
Audio streaming
Durable sync
Tool argument schema parity
Reasoning summaries
Prompt caching
Web search
File search
Code interpreter
Shell tool
Image generation
Computer use
MCP/connectors, unless separately enabled by AWS release status
Tool search, unless separately enabled by AWS release status
Skills or patch tools
Server-side tools
Vision
Audio input
Batches
Fine-tuning
Files
Tenancy
OpenAI Regions behavior
Troubleshooting
If a request fails, first confirm:
The selected OpenAI model is available in the AWS Region being used
The AWS account has access to the model in Amazon Bedrock
The AWS identity has the required IAM permissions
Quotas and rate limits are sufficient for the request
The request uses a capability currently supported by Amazon Bedrock
The request is being sent to the Amazon Bedrock endpoint, not the OpenAI-hosted API endpoint
Support
For issues related to AWS authentication, IAM permissions, Bedrock model access, quotas, billing, regional availability, Bedrock request failures, AWS backend logs, or Bedrock service behavior, contact AWS Support.
For information about OpenAI’s hosted Responses API, see OpenAI’s API documentation. OpenAI’s hosted API documentation describes OpenAI Platform behavior and may include capabilities that are not available in the Amazon Bedrock implementation.