Openai Models Amazon Bedrock Governed Responses Poc 2026

• Date: 2026-07-08T00:39:28.542000+00:00
• Model: gpt-5.5-2026-04-23
• Purpose: baseline Responses API smoke test to anchor a Bedrock migration readiness matrix
• Request ID: 8f6b3c13-4dbd-43d5-bef1-cff9e453f6f1
• Usage: {"input_tokens": 193, "input_tokens_details": {"cached_tokens": 0}, "output_tokens": 737, "output_tokens_details": {"reasoning_tokens": 516}, "total_tokens": 930}
• Prompt SHA-256: 08d826d197a2781205b6be58ac2e10e059e5ebcfd5255ebb6fa5b3f432a917eb
• Safety boundary: no confidential, customer, credential, or private data should be included in this run.

Prompt Excerpt

You are helping verify the request shape for an application that will move from the
standard OpenAI Responses API (https://api.openai.com/v1) to the Amazon Bedrock
"bedrock-mantle" OpenAI-compatible Responses endpoint, without changing application logic.

Task: In under 150 words, list the smallest set of code changes a developer must make to
point an existing Responses API call at Bedrock, assuming they only change configuration
(not business logic). Be precise and do not invent feature support. If a capability is not
part of the base Responses request/response contract, say so plainly.

This is a non-confidential lab check. Do not include any secrets or private data.

Output

Smallest config-only changes:

1. Change the OpenAI client base URL from `https://api.openai.com/v1` to the Bedrock OpenAI-compatible base URL for your AWS Region, e.g. Bedrock’s `/openai/v1` endpoint.

2. Replace OpenAI authentication with the Bedrock-supported authentication method for that endpoint. Do not send OpenAI org/project headers unless Bedrock explicitly documents them.

3. Change the `model` value from an OpenAI model name to the Bedrock model ID / inference profile / provisioned endpoint name that Bedrock expects, e.g. the configured `bedrock-mantle` target.

4. Keep the Responses call shape the same only for the base contract: `POST /responses` with fields such as `model` and `input`, and parse the standard Responses response fields your app already uses.

Do not assume support for non-base OpenAI capabilities such as hosted tools, file search, code interpreter, custom OpenAI headers, or OpenAI-specific tracing/metadata unless Bedrock documents them.

Limitations

This API run is a bounded lab check. It is not a production benchmark, user study, or proof that an external product works in a real customer environment.

What this run does and does not prove (Effloow Lab scope)

What was actually executed. One live call to the standard OpenAI Responses API (POST https://api.openai.com/v1/responses, model gpt-5.5-2026-04-23), completed status, request id 8f6b3c13-4dbd-43d5-bef1-cff9e453f6f1. The usage object came back as input_tokens=193, output_tokens=737 (of which reasoning_tokens=516), total_tokens=930, cached_tokens=0. This anchors the exact request/response contract a reader would keep when repointing the same call at Bedrock.

What was NOT executed. No call was made to the Amazon Bedrock bedrock-mantle endpoint. Effloow does not have an AWS account configured for this run, so live Bedrock behavior, per-region model availability, latency versus the direct OpenAI endpoint, and CloudWatch AWS/BedrockMantle metric emission are all [DATA NOT AVAILABLE] and are cited from official documentation only.

Source-derived readiness matrix (documentation, not measured)

Documented UNSUPPORTED features on Bedrock (from OpenAI's Bedrock guide)

Audio input; WebSocket connections; hosted web search; hosted file search; computer use; shell tool; image generation tool; remote MCP servers.

Sources: OpenAI "OpenAI models on Amazon Bedrock" guide; OpenAI API changelog entry 2026-06-01; AWS What's New "CloudWatch metrics for the bedrock-mantle endpoint" 2026-06-01; AWS What's New "redesigned Bedrock console" 2026-06-04.