GPT-6.1 Terra: AWS Bedrock Setup and Tool Calling

One-sentence takeaway

To use GPT-6.1 Terra on Bedrock with server-side tool calling, choose bedrock-mantle, which supports that capability. The model card's standard request examples use the Responses API, but it does not show the API request for tool calling.

Use cases

• Suitable tasks: Calling GPT-6.1 Terra on Amazon Bedrock with the OpenAI Python SDK, choosing an endpoint and model identifier, and determining whether server-side tool calling is supported.

• Unsuitable tasks: Configuring a tool schema or reproducing a complete tool call based only on this model card; the page provides no tool declaration, call parameters, or return example.

• Applicable model version: GPT-6.1 Terra. The Bedrock model ID is openai.gpt-6.1-terra; for cross-region calls through bedrock-runtime, use an inference profile ID instead.

• Applicable client, agent, or API: OpenAI Python SDK. Both bedrock-mantle and bedrock-runtime support Responses and Chat Completions. The model card recommends using bedrock-runtime for new applications where possible, but this endpoint does not support server-side tool calling for this model.

• Recommended reasoning tier and parameters: Not specified. The model card does not list reasoning tiers or recommended reasoning parameters for Terra on Bedrock.

Ready-to-use content

First install the SDK used in the official example, then create a long-term API key in the Amazon Bedrock console:

pip install openai

The model card provides configurations for two endpoints. Replace the placeholder with your Bedrock API key; do not commit the key to a code repository.

bedrock-mantle:

OPENAI_API_KEY="<Bedrock API key>"
OPENAI_BASE_URL="https://bedrock-mantle.us-east-1.api.aws/openai/v1"

from openai import OpenAI

client = OpenAI()
response = client.responses.create(
    model="openai.gpt-6.1-terra",
    input="Can you explain the features of Amazon Bedrock?"
)
print(response)

bedrock-runtime:

OPENAI_API_KEY="<Bedrock API key>"
OPENAI_BASE_URL="https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1"

from openai import OpenAI

client = OpenAI()
response = client.responses.create(
    model="us.openai.gpt-6.1-terra",
    input="Can you explain the features of Amazon Bedrock?"
)
print(response)

When using bedrock-runtime, the page requires a cross-region inference profile ID in the request, such as us.openai.gpt-6.1-terra or global.openai.gpt-6.1-terra; this endpoint does not offer in-region inference for this model. In addition to the inference profile, the IAM identity needs bedrock:InvokeModel permission for the account's default project.

Test or workflow steps

1. Generate a long-term API key in the Bedrock console and install the openai Python package.

2. Set OPENAI_API_KEY and OPENAI_BASE_URL for the target endpoint.

3. For bedrock-mantle, use openai.gpt-6.1-terra. For bedrock-runtime, use an available cross-region inference profile ID and confirm that IAM also allows access to the default project.

4. Send a basic request with the Responses API and inspect the response.

5. If the application needs server-side tool calling, choose bedrock-mantle. If you choose bedrock-runtime, do not rely on server-side tool calling. Configure the tool schema and complete call flow using the relevant API documentation; this page does not provide those details.

flowchart TD
    A[Create a Bedrock API key and install the OpenAI SDK] --> B{Choose an endpoint}
    B -->|Need server-side tool calling| C[bedrock-mantle; configure the API using the tool-calling documentation]
    B -->|Standard Responses call| D[bedrock-runtime + cross-region inference ID]
    C --> E[Verify the response]
    D --> F[Check default-project IAM permissions]
    F --> E

Raw evidence and data

• The AWS model card lists two supported endpoints: bedrock-mantle and bedrock-runtime. Both support Responses and Chat Completions. The model card also lists Converse and Invoke support for bedrock-runtime; bedrock-mantle does not support those two APIs.

• The bedrock-mantle model ID is openai.gpt-6.1-terra, and its endpoint format is https://bedrock-mantle.{region}.api.aws/openai/v1; the page's example uses us-east-1.

• The bedrock-runtime model ID is openai.gpt-6.1-terra, but calls must specify a cross-region inference profile ID. Its base URL format is https://bedrock-runtime.{region}.amazonaws.com/openai/v1.

• The model card lists server-side tool calling as supported on bedrock-mantle and server-side tool use as unsupported on bedrock-runtime.

• The model card's example uses the OpenAI Python SDK and the Responses API. The page does not show code for tool definitions, tool executors, or tool results.

• The IAM identity for bedrock-runtime needs bedrock:InvokeModel permission for both the inference profile and the account's default project.

Limitations

• Endpoint availability, regional support, and inference profiles may change; verify them against the current AWS console and your account configuration before deployment.

• us.openai.gpt-6.1-terra in the bedrock-runtime example is a cross-region inference profile identifier, not an in-region model ID. The page explicitly states that this endpoint does not support in-region inference for this model.

• The model card marks server-side tool calling as supported only on bedrock-mantle. The source provides no tool schema, request example, or call result, so this article does not add copyable tool-calling code.

• The model card recommends using bedrock-runtime for new applications where possible. If server-side tool calling is needed, choose bedrock-mantle based on the capability table and check the other endpoint limitations against the application's needs.

• This article records configuration from official documentation; it does not claim that the setup was run and verified in any specific AWS account, region, or IAM policy.

Reproduction notes

During collection, the AWS official model card and its public Markdown version were opened in the Tabbit browser. The endpoint/API support table, Bedrock feature table, programmatic access table, and Sample Code were checked. The two endpoint configurations and Responses API request were extracted; tool calling was recorded only as explicitly marked in the page's endpoint capability information. Collection date: 2026-10-27. AWS documentation and regional support information may change later.