> ## Documentation Index
> Fetch the complete documentation index at: https://quwan-b8010aef.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Responses API 思考模式

> 使用Responses API启用思考模式的示例，展示模型的推理过程

# Responses API 思考模式

思考模式允许模型展示其推理过程，帮助用户理解AI是如何得出答案的。gpt-5 模型支持原生的思考推理模式。

> **重要提示**：根据内容安全管理政策，必须同时设置 `"effort"` 和 `"summary": "detailed"` 参数才能开启思考过程展示功能。

## 快速开始

只需要替换 `<API-KEY>` 为你的实际API密钥即可运行。

```bash cURL theme={null}
curl -X POST "https://api.tokenops.ai/v1/responses" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <API-KEY>" \
  -d '{
    "model": "gpt-5-2025-08-07",
    "input": "请解决这个数学问题：一个圆的面积是78.54平方厘米，求这个圆的周长。请详细说明解题思路。",
    "reasoning": {
      "effort": "medium",
      "summary": "detailed"
    }
  }'
```

```python Python theme={null}
import requests
import json

# 配置API密钥和基础URL
API_KEY = "<API-KEY>"
BASE_URL = "https://api.tokenops.ai/v1"

def create_thinking_response():
    """使用gpt-5模型的思考模式"""
    url = f"{BASE_URL}/responses"
    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {API_KEY}"
    }

    payload = {
        "model": "gpt-5-2025-08-07",
        "input": "请解决这个数学问题：一个圆的面积是78.54平方厘米，求这个圆的周长。请详细说明解题思路。",
        "reasoning": {
            "effort": "medium",  # 推理强度：low, medium, high
            "summary": "detailed"  # 必需参数：开启思考过程展示
        }
    }

    response = requests.post(url, headers=headers, json=payload)

    if response.status_code == 200:
        result = response.json()

        # 获取推理过程（如果有）
        if "output" in result:
            for output_item in result["output"]:
                if output_item.get("type") == "reasoning" and "summary" in output_item:
                    print("=== 推理摘要 ===")
                    for summary_item in output_item["summary"]:
                        if summary_item.get("type") == "summary_text":
                            print(summary_item["text"])
                    print()

        # 获取最终回答
        if "output" in result:
            for output_item in result["output"]:
                if output_item.get("type") == "message" and "content" in output_item:
                    print("=== 最终回答 ===")
                    for content_item in output_item["content"]:
                        if content_item.get("type") == "output_text":
                            print(content_item["text"])

        # 显示使用统计
        usage = result.get("usage", {})
        reasoning_tokens = usage.get("output_tokens_details", {}).get("reasoning_tokens", 0)
        print(f"\nToken使用: {usage.get('total_tokens', 0)} (输入: {usage.get('input_tokens', 0)}, 输出: {usage.get('output_tokens', 0)}, 推理: {reasoning_tokens})")

        return result
    else:
        print(f"HTTP错误: {response.status_code}")
        print(response.text)
        return None

# 使用示例
if __name__ == "__main__":
    print("使用gpt-5模型的原生思考模式")
    response = create_thinking_response()
```

```javascript JavaScript/Node.js theme={null}
const axios = require('axios');

// 配置API密钥和基础URL
const API_KEY = '<API-KEY>';
const BASE_URL = 'https://api.tokenops.ai/v1';

async function createThinkingResponse() {
    const url = `${BASE_URL}/responses`;
    const headers = {
        'Content-Type': 'application/json',
        'Authorization': `Bearer ${API_KEY}`
    };

    const payload = {
        model: 'gpt-5-2025-08-07',
        input: '请解决这个数学问题：一个圆的面积是78.54平方厘米，求这个圆的周长。请详细说明解题思路。',
        reasoning: {
            effort: 'medium',
            summary: 'detailed'  // 必需参数：开启思考过程展示
        }
    };

    try {
        const response = await axios.post(url, payload, { headers });

        // 获取推理过程（如果有）
        if (response.data.output) {
            response.data.output.forEach(outputItem => {
                if (outputItem.type === 'reasoning' && outputItem.summary) {
                    console.log('=== 推理摘要 ===');
                    outputItem.summary.forEach(summaryItem => {
                        if (summaryItem.type === 'summary_text') {
                            console.log(summaryItem.text);
                        }
                    });
                    console.log();
                }
            });
        }

        // 获取最终回答
        if (response.data.output) {
            response.data.output.forEach(outputItem => {
                if (outputItem.type === 'message' && outputItem.content) {
                    console.log('=== 最终回答 ===');
                    outputItem.content.forEach(contentItem => {
                        if (contentItem.type === 'output_text') {
                            console.log(contentItem.text);
                        }
                    });
                }
            });
        }

        // 显示使用统计
        const usage = response.data.usage || {};
        const reasoningTokens = usage.output_tokens_details?.reasoning_tokens || 0;
        console.log(`\nToken使用: ${usage.total_tokens || 0} (输入: ${usage.input_tokens || 0}, 输出: ${usage.output_tokens || 0}, 推理: ${reasoningTokens})`);

        return response.data;
    } catch (error) {
        console.error('请求出错:', error.response?.data || error.message);
        return null;
    }
}

// 使用示例
async function main() {
    try {
        console.log('使用gpt-5模型的原生思考模式');
        await createThinkingResponse();
    } catch (error) {
        console.error('执行出错:', error);
    }
}

main();
```

```go Go theme={null}
package main

import (
    "bytes"
    "encoding/json"
    "fmt"
    "net/http"
)

const (
    API_KEY = "<API-KEY>"
    BASE_URL = "https://api.tokenops.ai/v1"
)

type ThinkingRequest struct {
    Model     string            `json:"model"`
    Input     string            `json:"input"`
    Reasoning *ReasoningOptions `json:"reasoning,omitempty"`
}

type ReasoningOptions struct {
    Effort  string `json:"effort,omitempty"`
    Summary string `json:"summary,omitempty"`
}

type ThinkingResponse struct {
    ID        string       `json:"id"`
    Object    string       `json:"object"`
    CreatedAt int64        `json:"created_at"`
    Status    string       `json:"status"`
    Model     string       `json:"model"`
    Output    []OutputItem `json:"output"`
    Usage     Usage        `json:"usage"`
    Reasoning *ReasoningInfo `json:"reasoning,omitempty"`
}

type OutputItem struct {
    ID      string         `json:"id"`
    Type    string         `json:"type"`
    Summary []SummaryItem  `json:"summary,omitempty"`
    Content []ContentItem  `json:"content,omitempty"`
    Status  string         `json:"status,omitempty"`
    Role    string         `json:"role,omitempty"`
}

type SummaryItem struct {
    Type string `json:"type"`
    Text string `json:"text"`
}

type ContentItem struct {
    Type string `json:"type"`
    Text string `json:"text"`
}

type Usage struct {
    InputTokens  int `json:"input_tokens"`
    OutputTokens int `json:"output_tokens"`
    TotalTokens  int `json:"total_tokens"`
}

type ReasoningInfo struct {
    Effort  string `json:"effort"`
    Summary string `json:"summary"`
}

func createThinkingResponse() (*ThinkingResponse, error) {
    url := BASE_URL + "/responses"

    request := ThinkingRequest{
        Model: "gpt-5-2025-08-07",
        Input: "请解决这个数学问题：一个圆的面积是78.54平方厘米，求这个圆的周长。请详细说明解题思路。",
        Reasoning: &ReasoningOptions{
            Effort:  "medium",
            Summary: "detailed", // 必需参数：开启思考过程展示
        },
    }

    jsonData, err := json.Marshal(request)
    if err != nil {
        return nil, err
    }

    httpReq, err := http.NewRequest("POST", url, bytes.NewBuffer(jsonData))
    if err != nil {
        return nil, err
    }

    httpReq.Header.Set("Content-Type", "application/json")
    httpReq.Header.Set("Authorization", "Bearer "+API_KEY)

    client := &http.Client{}
    resp, err := client.Do(httpReq)
    if err != nil {
        return nil, err
    }
    defer resp.Body.Close()

    var result ThinkingResponse
    err = json.NewDecoder(resp.Body).Decode(&result)
    if err != nil {
        return nil, err
    }

    // 显示推理摘要和最终回答
    for _, outputItem := range result.Output {
        if outputItem.Type == "reasoning" && len(outputItem.Summary) > 0 {
            fmt.Println("=== 推理摘要 ===")
            for _, summaryItem := range outputItem.Summary {
                if summaryItem.Type == "summary_text" {
                    fmt.Println(summaryItem.Text)
                }
            }
            fmt.Println()
        }
        
        if outputItem.Type == "message" && len(outputItem.Content) > 0 {
            fmt.Println("=== 最终回答 ===")
            for _, contentItem := range outputItem.Content {
                if contentItem.Type == "output_text" {
                    fmt.Println(contentItem.Text)
                }
            }
        }
    }
    fmt.Printf("\nToken使用: %d (输入: %d, 输出: %d)\n",
        result.Usage.TotalTokens,
        result.Usage.InputTokens,
        result.Usage.OutputTokens)

    return &result, nil
}

func main() {
    fmt.Println("使用gpt-5模型的原生思考模式")
    response, err := createThinkingResponse()
    if err != nil {
        fmt.Printf("错误: %v\n", err)
    }

    if response != nil {
        fmt.Printf("\n响应ID: %s\n", response.ID)
    }
}
```

## gpt-5 模型的思考模式实现

gpt-5 模型内置推理能力，通过 `reasoning` 参数控制推理强度和思考过程展示：

```json theme={null}
{
  "model": "gpt-5-2025-08-07",
  "input": "your question here",
  "reasoning": {
    "effort": "medium",  // low, medium, high
    "summary": "detailed"  // 必需参数：开启思考过程展示
  }
}
```

响应会包含 `reasoning` 字段，展示模型的推理摘要。

## 响应结果示例

```json theme={null}
{
  "id": "resp_01be82131ec287250068fdf7cd8d6c81969af11dab7e4218fe",
  "object": "response",
  "created_at": 1761474509,
  "status": "completed",
  "model": "gpt-5-2025-08-07",
  "output": [
    {
      "id": "rs_01be82131ec287250068fdf7ce17a081968ba0933d6ff8baf0",
      "type": "reasoning",
      "summary": [
        {
          "type": "summary_text",
          "text": "**Calculating circle properties**\n\nI'm laying out the steps for calculating the radius and circumference of a circle with an area of 78.54 cm². I'll use the formula A = πr², which gives me the radius, r = sqrt(A/π). So, with A plugged in, I find r to be 5 cm. \n\nNext, I calculate the circumference as C = 2πr, getting about 31.416 cm. I'll also explore an alternative formula for circumference, but it involves a lot of detailed computation that I just need to clarify and keep organized."
        },
        {
          "type": "summary_text",
          "text": "**Reviewing calculations for circumference**\n\nI'm comparing earlier calculations where I found the product Aπ to be 246.3, but the correct value should be 246.741264. Then, using 2√(Aπ), I calculate 2 * 15.712, which gives me about 31.424. This is slightly different from the exact value of 31.416 using r = 5 cm and π = 3.1416. \n\nI realize that the discrepancies in rounding might affect the results, and I'm clarifying that while A = 78.54 is technically equal to 25 * 3.1416, the algebraic relations hold true if I use the same π value consistently."
        }
      ]
    },
    {
      "id": "msg_01be82131ec287250068fdf7e0476c8196b7f3201011b0453d",
      "type": "message",
      "status": "completed",
      "content": [
        {
          "type": "output_text",
          "text": "解题思路：\n- 用到两个公式：\n  1) 面积公式 A = πr²\n  2) 周长公式 C = 2πr\n- 由面积求半径：r = √(A/π)\n  代入 A = 78.54，取 π ≈ 3.1416，\n  r = √(78.54 / 3.1416) = √25 = 5（厘米）\n- 求周长：C = 2πr = 2 × 3.1416 × 5 = 31.416（厘米）\n\n结论：该圆的周长约为31.416厘米（可记为31.42厘米）。此外，可写成 C = 10π 厘米。"
        }
      ],
      "role": "assistant"
    }
  ],
  "reasoning": {
    "effort": "medium",
    "summary": "detailed"
  },
  "usage": {
    "input_tokens": 38,
    "input_tokens_details": {
      "cached_tokens": 0
    },
    "output_tokens": 2348,
    "output_tokens_details": {
      "reasoning_tokens": 2176
    },
    "total_tokens": 2386
  }
}
```

## 参数说明

### 推理参数

* `reasoning.effort` - 推理强度
  * `"low"` - 快速推理
  * `"medium"` - 平衡推理（推荐）
  * `"high"` - 深度推理
* `reasoning.summary` - 思考过程展示控制（**必需参数**）
  * `"detailed"` - 展示详细的思考过程和推理步骤

> **注意**：根据内容安全管理政策，必须设置 `"summary": "detailed"` 才能开启思考模式功能。

### 通用参数

* `max_tokens` - 最大生成token数（建议2000以上用于思考模式）
* `temperature` - 控制输出的随机性

## 应用场景

1. **数学问题解决** - 展示计算步骤和推理过程
2. **逻辑推理** - 显示逻辑分析的每个环节
3. **复杂决策** - 说明考虑的因素和权衡过程
4. **代码调试** - 展示问题分析和解决思路
5. **学术研究** - 提供详细的论证过程

## 注意事项

1. **必需参数**: 必须设置 `"summary": "detailed"` 参数才能开启思考模式
2. **Token消耗**: 思考模式通常消耗更多token，包括推理token
3. **响应时间**: 启用思考可能增加响应时间
4. **内容质量**: 思考过程有助于提高回答质量和可信度
5. **隐私考虑**: 思考内容可能包含敏感的推理过程

## 最佳实践

1. **问题类型**: 复杂问题更适合使用思考模式
2. **参数设置**: 根据问题复杂度选择合适的reasoning.effort
3. **必需配置**: 始终包含 `"summary": "detailed"` 参数
4. **结果解析**: 合理处理思考内容和最终答案的分离
5. **Token管理**: 考虑思考模式的额外token消耗


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.