AnyoneLearnAI
Start

Reference · API

AWS Bedrock invoke model

Last updated

Call **foundation models inside AWS** — Claude, Llama, Titan, and others — with IAM auth instead of vendor API keys in app code.

Call **foundation models inside AWS** — Claude, Llama, Titan, and others — with IAM auth instead of vendor API keys in app code.

#When to use

  • Workloads already on AWS with strict VPC / compliance needs
  • Centralized model access via IAM roles and CloudTrail
  • Swapping model IDs without rewriting client SDKs (within Bedrock catalog)

    • Conceptual pattern — exact model IDs and request fields vary by provider on Bedrock.

    #Endpoint (conceptual)

    POST https://bedrock-runtime.{region}.amazonaws.com/model/{modelId}/invoke

    Anthropic-on-Bedrock uses a messages-style body; others use provider-specific JSON.

    #Request shape (sketch)

    {
  "anthropic_version": "bedrock-2023-05-31",
  "max_tokens": 512,
  "messages": [{"role": "user", "content": "Summarize this ticket."}]
}

    Sign with **AWS Signature V4** (SDK handles this).

    #Python sketch (boto3)

    import boto3, json
client = boto3.client("bedrock-runtime", region_name="us-east-1")
body = {
    "anthropic_version": "bedrock-2023-05-31",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Hello"}],
}
resp = client.invoke_model(
    modelId="anthropic.claude-3-5-sonnet-20241022-v2:0",
    body=json.dumps(body),
)
print(json.loads(resp["body"].read()))

    #Common pitfalls

    | Pitfall | Fix |

    |---------|-----|

    | Wrong region for model ID | Enable model access in Bedrock console per region |

    | Copy-paste direct API bodies | Bedrock wraps Anthropic/OpenAI shapes differently |

    | No IAM least privilege | Scope `bedrock:InvokeModel` to specific model ARNs |

    | Ignoring latency / quotas | Request quota increases; use async for batch |

    Learn by doing → Try the interactive lesson →
    ← All API docs