Events Reference

This page lists all Server-Sent Events (SSE) emitted by the streaming API.

Response Lifecycle

Events that describe the state of the overall response.

`response.created`

Fired immediately when the connection is established.

Payload: { "response": { "id": "...", "status": "in_progress", ... } }

`response.in_progress`

Fired periodically or after major state changes.

Payload: { "response": { ... } }

`response.completed`

Fired when the response generation is successfully finished.

Payload: { "response": { "status": "completed", "output": [...], "usage": {...} } }

`response.incomplete`

Fired when generation reaches an incomplete terminal state (for example output-token truncation or content filtering).

Payload: { "response": { "status": "incomplete", "incomplete_details": {"reason": "max_output_tokens"}, ... } }

`response.failed`

Fired if an error occurs during generation.

Payload: { "response": { "status": "failed", "error": {...} } }

Output Items

Events related to the creation and completion of items in the output list (messages, tool calls).

`response.output_item.added`

Fired when a new item (message or tool call) is initialized.

Payload:

{
  "output_index": 0,
  "item": { "type": "message", "id": "...", "status": "in_progress" }
}

`response.output_item.done`

Fired when an item is fully generated.

Payload:

{
  "output_index": 0,
  "item": { "status": "completed", "content": [...] }
}

Content Parts

Events for parts within an output item (e.g., text blocks within a message).

`response.content_part.added`

Fired when a new content part begins.

Payload: { "output_index": 0, "content_index": 0, "part": { "type": "text" } }

`response.content_part.done`

Fired when a content part is finished.

Payload: { "output_index": 0, "content_index": 0, "part": { ... } }

Text Generation

`response.output_text.delta`

Fired when new text tokens are generated.

Payload: { "delta": "...", "output_index": 0, "content_index": 0 }

`response.output_text.done`

Fired when the text block is complete.

Payload: { "text": "Full text...", ... }

Tool Calls

Function Calling (Custom)

`response.function_call_arguments.delta`

Streaming JSON arguments for a function call.

Payload: { "delta": "{\"loc\":", ... }

`response.function_call_arguments.done`

Final arguments string.

Payload: { "arguments": "{\"loc\":\"Paris\"}", ... }

Code Interpreter (Built-in)

`response.code_interpreter_call.in_progress`

The tool call has started.

`response.code_interpreter_call_code.delta`

The model is writing the Python code to be executed.

Payload: { "delta": "print(..." }

`response.code_interpreter_call_code.done`

The code block is complete.

`response.code_interpreter_call.interpreting`

The code has been sent to the runtime and is executing.

`response.code_interpreter_call.completed`

Execution has finished (logs/images are available in the final item).

MCP Tool Calls (Hosted + Client-Specified Remote)

`response.mcp_call.in_progress`

MCP call has started.

`response.mcp_call_arguments.delta`

Streaming MCP arguments JSON chunks.

Payload: { "delta": "{\"query\":", ... }

`response.mcp_call_arguments.done`

Final MCP arguments JSON string.

Payload: { "arguments": "{\"query\":\"migration notes\"}", ... }

`response.mcp_call.completed`

MCP call completed successfully.

Payload: { "output": "{...}", ... }

`response.mcp_call.failed`

MCP call failed at item level.

Payload: { "item_id": "...", "output_index": 2, "sequence_number": 53, ... }

When response.output_item.done is emitted for MCP, the item type is mcp_call and includes:

server_label
name
arguments
status (completed or failed)
output (on success) or error (on failure)

Reasoning

For models that support reasoning (Chain of Thought).

`response.reasoning.delta`

Streaming reasoning content.

`response.reasoning.done`

Reasoning is complete.