A list containing all the conversations between the user and the assistant. Each item in the list should be a dictionary with two keys: 'role' and 'message'.

role: Specifies the role of the speaker and can have the values 'user', 'system', 'assistant' or 'tool'. The system role instructs the way the model should answer, e.g. 'You are a helpful assistant'. The user role specifies the user query and assistant is the model's response. The tool role is for external tools that can be used in the conversation.

message: A list of dictionaries. Each dictionary in the 'message' list must contain the keys 'type' and 'content'.

Structure

• type: Specifies the type of content and can be 'image_url' or 'text'.
• content: A dictionary with the actual content based on the 'type':
  • If 'type' is 'image_url', 'content' must contain 'image_url' and must not contain 'text'.
  • If 'type' is 'text', 'content' must contain 'text' and must not contain 'image_url'.

Example

[
  {
    "role": "user",
    "content": [
        {
          "type": "text",
          "text": "Describe this image"
        },
        {
        "type": "image_url",
        "image_url": {
          "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"
          }
        }
    ]
  }
]

Optional list of metadata associated with the chat request. Can be used to provide additional context or tracking information.

Example:

{
  "metadata": [
    {"key": "conversation_id", "value": "chat_12345"},
    {"key": "source", "value": "customer_support"}
  ]
}

Modify the likelihood of specific tokens appearing in the response. A dictionary where keys are token IDs and values are bias scores.

Example:

{
  "logit_bias": {
    "50256": -100,  # Reduce probability of end-of-text token
    "15": 5  # Slightly increase probability of a specific token
  }
}

List of supported input/output modalities for the chat.

Example:

{
  "modalities": ["text", "image", "audio"]
}

Optional field for storing prediction-related information. Flexible dictionary to capture model's predictive metadata.

Example:

{
  "prediction": {
    "confidence_score": 0.85,
    "top_prediction": "response_category"
  }
}

Optional dictionary for audio-related parameters or metadata.

Example:

{
  "audio": {
    "language": "en-US",
    "transcription_format": "srt"
  }
}

Specify the desired response format for the completion.

Example:

{
  "response_format": {
    "type": "json_object",
    "schema": {...}
  }
}

List of strings that will cause the model to stop generating.

Example:

{
  "stop": ["\n", "Human:", "AI:"]
}

Additional configuration for streaming responses.

Example:

{
  "stream_options": {
    "include_usage": true
  }
}

List of tools or function definitions available to the model.

Example:

{
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Retrieve current weather"
      }
    }
  ]
}

List of function definitions available to the model.

Example:

{
  "functions": [
    {
      "name": "get_current_weather",
      "description": "Get the current weather for a location",
      "parameters": {...}
    }
  ]
}

Configuration for enabling Claude's extended thinking. When enabled, responses include thinking content blocks showing Claude's thinking process before the final answer. Requires a minimum budget of 1,024 tokens and counts towards your max_tokens limit.

Example:

{
  'thinking': {
    'type': 'enabled'
    'budget_tokens': '1024'  }
}

Options for web search integration. Example: json web_search_options={ "search_context_size": "medium" # Options: "low", "medium", "high" }