[FEATURE REQ] Support for `response.reasoning_text.delta`

[anktsrkr]

Describe the feature or improvement you are requesting

As @stephentoub mentioned in microsoft/agent-framework#1250 currently OpenAI sdk does not support response.reasoning_text.delta which is emitted by gpt-oss, Ref : https://cookbook.openai.com/articles/gpt-oss/handle-raw-cot

SDK only supports response.reasoning_summary_text.delta for which I did not find a definitive statement that GPT-OSS emits such events.

In such case could we add this event as well?

Additional context

No response

[jsquire]

Hi @anktsrkr. Thanks for reaching out and for your request. The response.reasoning_text.data even is a newer event and is not yet represented in the set of response events in the REST API spec used by the client. Once the spec is updated to include this area, it will be added to the client.

In the meantime, you can use following approach to get access to the raw event stream and use SseParser to capture the event:

var client = new OpenAIResponseClient("gpt-4.1", "<< YOUR API KEY >>");

// Generally, samples show iterating over this collection to process each update.
AsyncCollectionResult<StreamingResponseUpdate> responseUpdates = client.CreateResponseStreamingAsync("Say 'this is a test.'");

// Instead, unwrap into each raw page.
await foreach (var page in responseUpdates.GetRawPagesAsync())
{
    // The content stream is available on the raw response.  Note that this is unbuffered, so 
    // you cannot access via the "Content" property - you must use the stream.
    var sseContentStream =  page.GetRawResponse().ContentStream;
}

Writing out the raw content stream for that example results in something like:

event: response.created
data: {"type":"response.created","sequence_number":0,"response":{"id":"resp_686c3ec48ccc8199bd8c1397c6c0c7e306226590dfdad53f","object":"response","created_at":1751924420,"status":"in_progress","background":false,"error":null,"incomplete_details":null,"instructions":null,"max_output_tokens":null,"max_tool_calls":null,"model":"gpt-4.1-2025-04-14","output":[],"parallel_tool_calls":true,"previous_response_id":null,"reasoning":{"effort":null,"summary":null},"service_tier":"auto","store":true,"temperature":1.0,"text":{"format":{"type":"text"}},"tool_choice":"auto","tools":[],"top_logprobs":0,"top_p":1.0,"truncation":"disabled","usage":null,"user":null,"metadata":{}}}

event: response.in_progress
data: {"type":"response.in_progress","sequence_number":1,"response":{"id":"resp_686c3ec48ccc8199bd8c1397c6c0c7e306226590dfdad53f","object":"response","created_at":1751924420,"status":"in_progress","background":false,"error":null,"incomplete_details":null,"instructions":null,"max_output_tokens":null,"max_tool_calls":null,"model":"gpt-4.1-2025-04-14","output":[],"parallel_tool_calls":true,"previous_response_id":null,"reasoning":{"effort":null,"summary":null},"service_tier":"auto","store":true,"temperature":1.0,"text":{"format":{"type":"text"}},"tool_choice":"auto","tools":[],"top_logprobs":0,"top_p":1.0,"truncation":"disabled","usage":null,"user":null,"metadata":{}}}

event: response.output_item.added
data: {"type":"response.output_item.added","sequence_number":2,"output_index":0,"item":{"id":"msg_123456789012345678901234567890123456789012345678","type":"message","status":"in_progress","content":[],"role":"assistant"}}

event: response.content_part.added
data: {"type":"response.content_part.added","sequence_number":3,"item_id":"msg_123456789012345678901234567890123456789012345678","output_index":0,"content_index":0,"part":{"type":"output_text","annotations":[],"logprobs":[],"text":""}}

event: response.output_text.delta
data: {"type":"response.output_text.delta","sequence_number":4,"item_id":"msg_123456789012345678901234567890123456789012345678","output_index":0,"content_index":0,"delta":"This","logprobs":[]}

event: response.output_text.delta
data: {"type":"response.output_text.delta","sequence_number":5,"item_id":"msg_123456789012345678901234567890123456789012345678","output_index":0,"content_index":0,"delta":" is","logprobs":[]}

event: response.output_text.delta
data: {"type":"response.output_text.delta","sequence_number":6,"item_id":"msg_123456789012345678901234567890123456789012345678","output_index":0,"content_index":0,"delta":" a","logprobs":[]}

event: response.output_text.delta
data: {"type":"response.output_text.delta","sequence_number":7,"item_id":"msg_123456789012345678901234567890123456789012345678","output_index":0,"content_index":0,"delta":" test","logprobs":[]}

event: response.output_text.delta
data: {"type":"response.output_text.delta","sequence_number":8,"item_id":"msg_123456789012345678901234567890123456789012345678","output_index":0,"content_index":0,"delta":".","logprobs":[]}

event: response.output_text.done
data: {"type":"response.output_text.done","sequence_number":9,"item_id":"msg_123456789012345678901234567890123456789012345678","output_index":0,"content_index":0,"text":"This is a test.","logprobs":[]}

event: response.content_part.done
data: {"type":"response.content_part.done","sequence_number":10,"item_id":"msg_123456789012345678901234567890123456789012345678","output_index":0,"content_index":0,"part":{"type":"output_text","annotations":[],"logprobs":[],"text":"This is a test."}}

event: response.output_item.done
data: {"type":"response.output_item.done","sequence_number":11,"output_index":0,"item":{"id":"msg_123456789012345678901234567890123456789012345678","type":"message","status":"completed","content":[{"type":"output_text","annotations":[],"logprobs":[],"text":"This is a test."}],"role":"assistant"}}

event: response.completed
data: {"type":"response.completed","sequence_number":12,"response":{"id":"resp_686c3ec48ccc8199bd8c1397c6c0c7e306226590dfdad53f","object":"response","created_at":1751924420,"status":"completed","background":false,"error":null,"incomplete_details":null,"instructions":null,"max_output_tokens":null,"max_tool_calls":null,"model":"gpt-4.1-2025-04-14","output":[{"id":"msg_123456789012345678901234567890123456789012345678","type":"message","status":"completed","content":[{"type":"output_text","annotations":[],"logprobs":[],"text":"This is a test."}],"role":"assistant"}],"parallel_tool_calls":true,"previous_response_id":null,"reasoning":{"effort":null,"summary":null},"service_tier":"default","store":true,"temperature":1.0,"text":{"format":{"type":"text"}},"tool_choice":"auto","tools":[],"top_logprobs":0,"top_p":1.0,"truncation":"disabled","usage":{"input_tokens":14,"input_tokens_details":{"cached_tokens":0},"output_tokens":6,"output_tokens_details":{"reasoning_tokens":0},"total_tokens":20},"user":null,"metadata":{}}}

[jsquire]

similar: #643.

[anktsrkr]

Thanks @jsquire for looking into this. I think here is the API spec for that -
https://platform.openai.com/docs/api-reference/responses-streaming/response/reasoning_text

My actual implementation was with agent framework as mentioned earlier and looking at the raw response, I can see it is coming as UnknownResponseStreamEvent

so, probably i need to play with that.