> ## Documentation Index
> Fetch the complete documentation index at: https://gomodel.enterpilot.io/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Amazon Bedrock Mantle

> Use Amazon Bedrock's OpenAI-compatible Mantle endpoint with direct Responses API routing, including GPT-5.6 Sol, Terra, and Luna.

Bedrock Mantle is Amazon Bedrock's OpenAI-compatible inference endpoint.
GoModel calls its Responses API directly, so Responses-only models are not
translated to Chat Completions or Bedrock Converse.

<Note>
  This is a separate provider type from [Amazon Bedrock Runtime](/docs/providers/bedrock).
  Use `bedrock-mantle` for OpenAI-compatible Mantle models and `bedrock` for
  Bedrock Runtime Converse models. You can configure both at the same time.
</Note>

## Prerequisites

Enable the models in your AWS account and choose a supported AWS Region. You
can authenticate with an Amazon Bedrock API key or the standard AWS credential
chain.

## Configure

<Tabs>
  <Tab title="Bedrock API key">
    ```bash theme={null}
    BEDROCK_MANTLE_API_KEY=ABSK...
    BEDROCK_MANTLE_BASE_URL=us-east-1
    GOMODEL_MASTER_KEY=change-me
    ```
  </Tab>

  <Tab title="AWS credentials (SigV4)">
    ```bash theme={null}
    BEDROCK_MANTLE_BASE_URL=us-east-1
    AWS_ACCESS_KEY_ID=...
    AWS_SECRET_ACCESS_KEY=...
    # AWS_SESSION_TOKEN=...       # temporary credentials only
    GOMODEL_MASTER_KEY=change-me
    ```
  </Tab>
</Tabs>

For SigV4, GoModel uses the AWS SDK credential chain, including
`AWS_PROFILE`, IAM Identity Center, and container or instance roles. In
production, prefer workload credentials over long-lived access keys.

`AWS_BEARER_TOKEN_BEDROCK` is also accepted as a bearer-token fallback. Set
`BEDROCK_MANTLE_BASE_URL` with it so provider discovery can enable Mantle.

The base URL can be an AWS Region or a complete endpoint. A region is the
recommended form:

```bash theme={null}
BEDROCK_MANTLE_BASE_URL=us-west-2
# Equivalent host: https://bedrock-mantle.us-west-2.api.aws
```

For an API-key or YAML configuration where the base URL is omitted, GoModel
uses `BEDROCK_MANTLE_REGION`, then `AWS_REGION`, then `AWS_DEFAULT_REGION`, and
finally `us-east-1`. With only IAM credentials, set
`BEDROCK_MANTLE_BASE_URL` so provider discovery enables Mantle.

Or configure the provider in `config.yaml`:

```yaml theme={null}
providers:
  bedrock-mantle:
    type: bedrock-mantle
    base_url: us-east-1
    api_key: "${BEDROCK_MANTLE_API_KEY}"
```

## Call GPT-5.6

```bash theme={null}
curl -s http://localhost:8080/v1/responses \
  -H "Authorization: Bearer $GOMODEL_MASTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai.gpt-5.6-sol",
    "input": "Explain why idempotency matters in distributed systems."
  }'
```

Streaming uses the same gateway endpoint with `"stream": true`.
`previous_response_id`, tools, reasoning controls, and unknown extension
fields are forwarded to Mantle without translating the request to chat.

### GPT-5.6 availability

| Model | Model ID               | Context | AWS Regions                           |
| ----- | ---------------------- | ------- | ------------------------------------- |
| Sol   | `openai.gpt-5.6-sol`   | 272K    | `us-east-1`, `us-east-2`              |
| Terra | `openai.gpt-5.6-terra` | 272K    | `us-east-1`, `us-east-2`, `us-west-2` |
| Luna  | `openai.gpt-5.6-luna`  | 272K    | `us-east-1`, `us-east-2`, `us-west-2` |

These models support the Responses API, not Chat Completions, Converse, or
Invoke. GoModel automatically sends them to `/openai/v1/responses`, the model-
specific path required by AWS. Other Mantle models use `/v1/responses` or
`/v1/chat/completions` as appropriate.

## Endpoint path selection

`BEDROCK_MANTLE_API_MODE` controls which OpenAI-compatible path GoModel uses:

| Value      | Behavior                                                                                        |
| ---------- | ----------------------------------------------------------------------------------------------- |
| `auto`     | Default. Uses `/openai/v1` for known model families that require it and `/v1` for other models. |
| `openai`   | Forces `/openai/v1` for inference requests.                                                     |
| `standard` | Forces `/v1` for inference requests.                                                            |

The model catalog always uses `/v1/models`. Use an override when AWS adds a
model before GoModel's automatic path list is updated.

In YAML, set the same option with `api_mode`:

```yaml theme={null}
providers:
  mantle-preview:
    type: bedrock-mantle
    base_url: us-east-1
    api_mode: openai
    models:
      - openai.gpt-5.6-sol
```

## Multiple Mantle providers

Use suffixed environment variables for separate regions or accounts:

```bash theme={null}
BEDROCK_MANTLE_EAST_API_KEY=ABSK...
BEDROCK_MANTLE_EAST_BASE_URL=us-east-1

BEDROCK_MANTLE_WEST_API_KEY=ABSK...
BEDROCK_MANTLE_WEST_BASE_URL=us-west-2
```

This registers `bedrock-mantle-east` and `bedrock-mantle-west`.

## Troubleshooting

* **GPT-5.6 returns `404` or rejects the request** — use the
  `bedrock-mantle` provider type. The Bedrock Runtime provider uses Converse,
  which GPT-5.6 does not support.
* **`403` / access denied** — enable model access and verify that the API key
  or IAM principal can invoke Bedrock Mantle inference in that Region.
* **Provider is not discovered with `AWS_BEARER_TOKEN_BEDROCK` or an IAM
  role** — set `BEDROCK_MANTLE_BASE_URL` to enable the keyless provider.
* **A newly released model uses the wrong route** — set
  `BEDROCK_MANTLE_API_MODE=openai` or `standard` for that provider instance.

## References

* [Inference using the Responses API](https://docs.aws.amazon.com/bedrock/latest/userguide/bedrock-mantle.html)
* [GPT-5.6 Sol](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-sol.html)
* [GPT-5.6 Terra](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-terra.html)
* [GPT-5.6 Luna](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-luna.html)
* [API compatibility by model](https://docs.aws.amazon.com/bedrock/latest/userguide/models-api-compatibility.html)
* [Bedrock Mantle quotas](https://docs.aws.amazon.com/bedrock/latest/userguide/quotas-mantle.html)