How to Call GPT-6.1 Sol Through the Bedrock Mantle Endpoint

You can point the stock OpenAI SDK at Amazon Bedrock and call GPT-6.1 Sol by changing a base URL, a key and a model ID. By the end you will have a working Responses call and a Chat Completions call against the Bedrock Mantle endpoint in us-east-1, plus the bedrock-runtime variant that uses the US or global inference profiles.

GPT-6.1 Sol launched on Bedrock on September 29, 2026, with a 1M-token context window and up to 131,072 output tokens, per the model card.

Pick the Bedrock Mantle endpoint or bedrock-runtime

Bedrock exposes this model on two endpoints. They differ in model ID, base URL and which APIs they accept.

| | bedrock-mantle | bedrock-runtime | | --- | --- | --- | | Model ID | openai.gpt-6.1-sol | us.openai.gpt-6.1-sol or global.openai.gpt-6.1-sol | | Base URL | https://bedrock-mantle.us-east-1.api.aws/openai/v1 | https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1 | | APIs | Responses, Chat Completions | Responses, Chat Completions, Converse, Invoke | | Where it runs | In-Region, us-east-1 only | US or global cross-Region inference |

This guide uses Mantle first because it is the only in-Region option.

1. Generate a Bedrock API key

Open the Bedrock console, go to API keys, and create a long-term key. Copy it somewhere safe, because you will paste it into an environment variable next.

2. Install the OpenAI SDK and set the environment

python3 -m pip install openai

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

The OpenAI Python client reads both variables on its own, so your code needs no Bedrock-specific setup. Note the /openai/v1 path: on Mantle it replaces the usual /v1.

3. Smoke-test the endpoint with curl

curl "$OPENAI_BASE_URL/responses" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "openai.gpt-6.1-sol",
    "input": "Say hello in five words."
  }'

A JSON response means the key, Region and model access are working. If you get an error here, fix it before debugging your application.

4. Call GPT-6.1 Sol with the Responses API

from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="openai.gpt-6.1-sol",
    input="Can you explain the features of Amazon Bedrock?",
    max_output_tokens=512,
)
print(response.output_text)

This is the sample from the model card. Set max_output_tokens on every call, since output tokens count 10x against your quota (more on that below).

5. Or use Chat Completions

If your code already speaks Chat Completions, the same endpoint and key serve it under /openai/v1/chat/completions:

from openai import OpenAI

client = OpenAI()

response = client.chat.completions.create(
    model="openai.gpt-6.1-sol",
    messages=[{"role": "user", "content": "Can you explain the features of Amazon Bedrock?"}],
)
print(response.choices[0].message.content)

The client builds that path from OPENAI_BASE_URL, so nothing else changes.

6. Switch to bedrock-runtime with an inference profile

If you want cross-Region routing instead of in-Region, change the base URL and use a profile ID as the model:

export 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-sol",  # or "global.openai.gpt-6.1-sol"
    input="Can you explain the features of Amazon Bedrock?",
    max_output_tokens=512,
)
print(response.output_text)

Use us. to keep inference in US geographies and global. to let Bedrock route worldwide. Direct in-Region invocation is not supported on bedrock-runtime, and the source Region must be enabled for the profile you pick.

Gotchas

• Region. Mantle serves GPT-6.1 Sol from us-east-1 only. Anywhere else, use the runtime profiles.
• Standard tier only. Omit service_tier or set it to "default". Priority, Flex and Reserved are not supported.
• Price by route. Mantle and the US profile carry a 10% premium over global rates. Global is $2.00 input and $10.00 output per million tokens up to 272K input tokens, and requests above that pay long-context rates on the whole request. See the pricing table.
• No explicit prompt caching. The pricing table lists cache rates, but the model card says explicit prompt caching is not supported, so don't budget around it.
• Other gaps. Server-side system tools, intelligent prompt routing and the count tokens API are unsupported.
• Quota burn. Each output token consumes 10 tokens of quota.
• Structured outputs. JSON Schema output is documented for non-streaming bedrock-runtime calls. Don't assume it on Mantle.
• Keys. A long-term key is fine for a first call. For long-running services, OpenAI's Bedrock guide recommends the AWS credential chain instead.

Pick the route by Region, price and the APIs you need, because in production those three decide where your data is processed, what you pay and which features you can rely on.