Build an Anthropic Messages API Request

The Anthropic Messages API is the primary interface for all Claude models. Unlike the OpenAI Chat Completions API, the Messages API separates the system prompt into a dedicated top-level field rather than placing it in the messages array. This structural difference is significant: Claude uses the system field as a higher-priority instruction channel that the model is strongly trained to follow, making system-level constraints more reliable than embedding them as user messages. This example shows a complete Messages API request for a document analysis task with the system field, a user message containing a document, and generation parameters. The max_tokens field is required in the Anthropic API (unlike OpenAI where it is optional) — omitting it causes a validation error. The model field must use the exact model ID including the full version string like "claude-3-5-sonnet-20241022". Key differences from OpenAI: temperature range is 0.0–1.0 (not 0.0–2.0), there is no presence_penalty or frequency_penalty parameter, and streaming is enabled via stream: true rather than a separate streaming endpoint. The Anthropic API also supports extended thinking for claude-3-7-sonnet, which enables multi-step reasoning with visible thought traces.

{
  "model": "claude-3-5-sonnet-20241022",
  "max_tokens": 1024,
  "system": "You are an expert document analyst. Extract key information accurately and present it in a structured format.",
  "messages": [
    {
      "role": "user",
      "content": "Please summarize the key points from this meeting notes document and identify action items:\n\nMeeting: Q4 Planning Session\nDate: November 15, 2024\nAttendees: Alice, Bob, Carol\n\nDiscussion: The team reviewed Q3 results showing 15% growth. For Q4, priorities are: (1) launch new API features by Dec 1, (2) reduce support tickets by 20%, (3) complete security audit. Bob will lead the API work, Carol owns support improvements, Alice manages the audit."
    }
  ]
}