InvokeModel: The Foundation

InvokeModel is the original, "Low-level" API for Bedrock. It requires you to know exactly which JSON structure a specific model (like Llama or Titan) expects. While it's more work, it gives you 100% control over the specialized features of each model.

1. Setting up Boto3

You need the boto3 library.

pip install boto3

2. The Python Implementation

Every model has a different body structure. Below is an example for Amazon Titan.

import boto3
import json

# Use 'bedrock-runtime' to call models
client = boto3.client("bedrock-runtime", region_name="us-east-1")

model_id = "amazon.titan-text-express-v1"

# Model-specific payload
body = {
    "inputText": "Explain quantum physics in 2 sentences.",
    "textGenerationConfig": {
        "maxTokenCount": 512,
        "temperature": 0.5
    }
}

response = client.invoke_model(
    modelId=model_id,
    contentType="application/json",
    accept="application/json",
    body=json.dumps(body)
)

# Parse the response body stream
response_body = json.loads(response["body"].read())
print(response_body["results"][0]["outputText"])

---

3. Why InvokeModel is becoming "Legacy"

• If you want to switch from Titan to Claude, you have to rewrite the body and the response_body parsing logic because they use different keys (e.g., inputText vs messages).

---

4. Visualizing the Payload Cycle

graph LR
    Input[Text Prompt] --> B[Wrap in Model JSON: {inputText: ...}]
    B --> Call[invoke_model API]
    Call --> Res[Raw JSON: {results: [...]}]
    Res --> P[Parse specific keys]
    P --> Final[Answer string]

---

💡 Guidance for Learners

Use InvokeModel only if you need a specialized parameter that the standard API doesn't support (like specific image-generation settings for Stable Diffusion). For 99% of text tasks, use the Converse API (Lesson 2).

---

Summary

• InvokeModel is the low-level API for Bedrock.
• The Body structure changes for every model provider.
• You must read and parse the binary response stream manually.
• It provides maximum control but minimum portability.