OpenRouter reasoning field support

Expected Behavior

Currently we are using OpenRouter to access models, As OpenAI compatiable API.

Therefore, it would be great if it could be compatible with OpenRouter's API.

Current Behavior

What OpenRouter response:

{
    "id": "gen-1768528898-MKtgrOMD5dK6dq3iyOSh",
    "provider": "Google",
    "model": "google/gemini-3-flash-preview",
    "object": "chat.completion.chunk",
    "created": 1768528898,
    "choices": [
        {
            "index": 0,
            "delta": {
                "role": "assistant",
                "content": "",
                "reasoning": "**Deciding on Initial Action**\n\nI've decided to first query the ...",
                "reasoning_details": [
                    {
                        "index": 0,
                        "type": "reasoning.text",
                        "text": "**Deciding on Initial Action**\n\nI've decided to first query the ...",
                        "format": "google-gemini-v1"
                    }
                ],
                "annotations": []
            },
            "finish_reason": null,
            "native_finish_reason": null,
            "logprobs": null
        }
    ]
}

For doc: https://openrouter.ai/docs/guides/best-practices/reasoning-tokens

What current ChatResponse get:

{
    "metadata": {
        "id": "gen-1768528898-MKtgrOMD5dK6dq3iyOSh",
        "model": "google/gemini-3-flash-preview",
        "rateLimit": {
            "requestsLimit": 0,
            "requestsRemaining": 0,
            "requestsReset": "PT0S",
            "tokensRemaining": 0,
            "tokensLimit": 0,
            "tokensReset": "PT0S"
        },
        "usage": {
            "promptTokens": 0,
            "completionTokens": 0,
            "nativeUsage": {},
            "totalTokens": 0
        },
        "promptMetadata": [],
        "empty": false
    },
    "results": [
        {
            "metadata": {
                "finishReason": "",
                "contentFilters": [],
                "empty": true
            },
            "output": {
                "messageType": "ASSISTANT",
                "metadata": {
                    "role": "ASSISTANT",
                    "messageType": "ASSISTANT",
                    "refusal": "",
                    "finishReason": "",
                    "index": 0,
                    "annotations": [],
                    "id": "gen-1768528898-MKtgrOMD5dK6dq3iyOSh",
                    "reasoningContent": ""
                },
                "toolCalls": [],
                "media": [],
                "text": ""
            }
        }
    ],
    "result": {
        "metadata": {
            "finishReason": "",
            "contentFilters": [],
            "empty": true
        },
        "output": {
            "messageType": "ASSISTANT",
            "metadata": {
                "role": "ASSISTANT",
                "messageType": "ASSISTANT",
                "refusal": "",
                "finishReason": "",
                "index": 0,
                "annotations": [],
                "id": "gen-1768528898-MKtgrOMD5dK6dq3iyOSh",
                "reasoningContent": ""
            },
            "toolCalls": [],
            "media": [],
            "text": ""
        }
    }
}

Want to retrieve the data from the reasoning, but the currently returned data structure does not contain it.

Start by locating the OpenAI-compatible chat response mapping that produces ChatResponse, then compare its handling of content with OpenRouter's reasoning and reasoning_details fields. Confirm the reasoning text is retained in the returned response and add focused coverage for a streaming chunk shaped like the example.