2578 Protocol Comparison C1dff757

API Protocol Comparison Report

OpenAI Classic (Chat Completions) vs OpenAI Responses vs Anthropic Messages

1. Overview

2. Request Fields Comparison

2.1 Core Parameters

2.2 Tool/Function Calling

Tool Declaration Format Examples

OpenAI Classic:

{
  "type": "function",
  "function": {
    "name": "get_weather",
    "description": "Get current weather",
    "parameters": {
      "type": "object",
      "properties": {
        "location": {"type": "string"}
      },
      "required": ["location"]
    },
    "strict": true
  }
}

OpenAI Responses:

{
  "type": "function",
  "name": "get_weather",
  "description": "Get current weather",
  "parameters": {
    "type": "object",
    "properties": {
      "location": {"type": "string"}
    },
    "required": ["location"]
  },
  "strict": true
}

Anthropic Messages:

{
  "name": "get_weather",
  "description": "Get current weather",
  "input_schema": {
    "type": "object",
    "properties": {
      "location": {"type": "string"}
    },
    "required": ["location"]
  }
}

2.3 Message/Content Structure

Role Mapping

Content Block Types

2.4 Response Format/Structured Outputs

3. Response Fields Comparison

3.1 Top-Level Response Structure

3.2 Finish/Stop Reason Mapping

3.3 Usage Structure

3.4 Tool Call Response Format

OpenAI Classic:

{
  "choices": [{
    "message": {
      "role": "assistant",
      "content": null,
      "tool_calls": [{
        "id": "call_abc123",
        "type": "function",
        "function": {
          "name": "get_weather",
          "arguments": "{\"location\": \"Paris\"}"
        }
      }]
    },
    "finish_reason": "tool_calls"
  }]
}

OpenAI Responses:

{
  "output": [
    {
      "type": "function_call",
      "id": "fc_abc123",
      "call_id": "call_abc123",
      "name": "get_weather",
      "arguments": "{\"location\": \"Paris\"}"
    }
  ]
}

Anthropic Messages:

{
  "content": [{
    "type": "tool_use",
    "id": "toolu_abc123",
    "name": "get_weather",
    "input": {"location": "Paris"}
  }],
  "stop_reason": "tool_use"
}

4. Streaming Events Comparison

4.1 Event Format

4.2 Stream Event Types

OpenAI Classic Stream Events

• Each chunk has object: "chat.completion.chunk"
• choices[].delta contains incremental content
• choices[].delta.content for text
• choices[].delta.tool_calls for tool calls
• Final chunk may have choices[].finish_reason
• Ends with data: [DONE]

OpenAI Responses Stream Events

Anthropic Messages Stream Events

4.3 Delta Structures

OpenAI Classic Delta:

{
  "id": "chatcmpl-xxx",
  "object": "chat.completion.chunk",
  "choices": [{
    "index": 0,
    "delta": {"content": "Hello"},
    "finish_reason": null
  }]
}

OpenAI Responses Delta:

event: response.text.delta
data: {"type": "response.text.delta", "item_id": "item_xxx", "output_index": 0, "content_index": 0, "delta": "Hello"}

Anthropic Messages Delta:

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "Hello"}}

5. Error Response Comparison

5.1 Error Structure

OpenAI (both APIs):

{
  "error": {
    "message": "Error description",
    "type": "invalid_request_error",
    "param": "messages",
    "code": "invalid_type"
  }
}

Anthropic:

{
  "type": "error",
  "error": {
    "type": "invalid_request_error",
    "message": "Error description"
  }
}

5.2 Error Code Mapping

6. Multimodal Capabilities

6.1 Image Input Support

6.2 Image Generation

6.3 Audio Support

6.4 Document/File Support

7. Special Features

7.1 Extended Thinking

7.2 Caching

7.3 Stateful Conversations

8. Conversion Strategies

8.1 Lossless vs Lossy Conversions

Lossless mappings:

• Text content between all three APIs
• Basic tool declarations (name, description, schema)
• Role mappings (user, assistant)
• Temperature, max_tokens, stop sequences
• Image inputs (URL and base64)

Lossy mappings (require degradation):

• top_k (Anthropic → OpenAI): Drop or store in metadata
• seed (OpenAI → Anthropic): Drop or store in metadata
• n multiple completions (OpenAI → others): Not supported
• Presence/frequency penalty (OpenAI → others): Not supported
• logprobs (OpenAI → others): Not supported
• Native audio I/O (OpenAI → Anthropic): Not supported
• Extended thinking format differences

8.2 Recommended Degradation Strategies

1. Unsupported parameters: Store in metadata.unsupported_params for round-trip preservation
2. Capability gaps: Raise CapabilityNotSupportedError with details
3. Value range differences: Clamp values (e.g., temperature 0-2 → 0-1)
4. Format differences: Transform to equivalent format (e.g., tool schema nesting)

9. References

• OpenAI Chat Completions API
• OpenAI Responses API
• OpenAI Migration Guide
• Anthropic Messages API
• Anthropic Streaming