大模型服务平台 TokenHub 深度思考

一、功能说明

深度思考模型支持在生成最终答案前先进行推理，提升复杂任务的准确性和可解释性。

二、适用场景

·复杂代码生成、代码修复、代码重构。

·数学推导、逻辑分析、多步骤决策。

·复杂信息抽取后再综合归纳。

·需要更强稳定性和更少推理失误的任务。

三、开启/关闭深度思考

通过 thinking 参数控制是否开启思考模式。

·开启深度思考："thinking":{"type":"enabled"}

·关闭深度思考："thinking":{"type":"disabled"}

四、支持模型

五、调用示例：开启深度思考

注意：说明：

注意：请将 YOUR_API_KEY 替换为你创建的 API Key。

cURLPythonNode.jsJavaGocurl -X POST 'https://tokenhub.tencentmaas.com/v1/chat/completions' \\ -H 'Content-Type: application/json' \\ -H 'Authorization: Bearer YOUR_API_KEY' \\ -d '{ "model": "MODEL 参数值", "messages": [ {"role": "user", "content": "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？"} ], "thinking": {"type": "enabled"}, "stream": false }'from openai import OpenAIclient = OpenAI( api_key="YOUR_API_KEY", base_url="https://tokenhub.tencentmaas.com/v1",)response = client.chat.completions.create( model="MODEL 参数值", messages=[ {"role": "user", "content": "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？"}, ], extra_body={"thinking": {"type": "enabled"}},)# OpenAI SDK 不直接声明 reasoning_content 字段，需用 getattr 访问msg = response.choices[0].messageif hasattr(msg, "reasoning_content"): print("思考过程:", getattr(msg, "reasoning_content"))print("最终回答:", msg.content)import OpenAI from 'openai';const client = new OpenAI({ apiKey: 'YOUR_API_KEY', baseURL: 'https://tokenhub.tencentmaas.com/v1',});// Node.js SDK：thinking 字段直接展开到顶层const response = await client.chat.completions.create({ model: 'MODEL 参数值', messages: [ { role: 'user', content: '小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？' }, ], thinking: { type: 'enabled' },} as any);const msg: any = response.choices[0].message;if (msg.reasoning_content) console.log('思考过程:', msg.reasoning_content);console.log('最终回答:', msg.content);import okhttp3.*;import com.google.gson.Gson;import java.util.*;public class ThinkingChat { public static void main(String[] args) throws Exception { Map<String, Object> body = new HashMap<>(); body.put("model", "MODEL 参数值"); body.put("messages", List.of( Map.of("role", "user", "content", "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？") )); body.put("thinking", Map.of("type", "enabled")); Request request = new Request.Builder() .url("https://tokenhub.tencentmaas.com/v1/chat/completions") .header("Authorization", "Bearer YOUR_API_KEY") .post(RequestBody.create(new Gson().toJson(body), MediaType.parse("application/json"))) .build(); try (Response response = new OkHttpClient().newCall(request).execute()) { // 响应体中 message.reasoning_content 为思考过程，message.content 为最终回答 System.out.println(response.body().string()); } }}package mainimport ( "bytes" "encoding/json" "fmt" "io" "net/http")func main() { body, _ := json.Marshal(map[string]interface{}{ "model": "MODEL 参数值", "messages": []map[string]string{ {"role": "user", "content": "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？"}, }, "thinking": map[string]string{"type": "enabled"}, }) req, _ := http.NewRequest("POST", "https://tokenhub.tencentmaas.com/v1/chat/completions", bytes.NewBuffer(body)) req.Header.Set("Authorization", "Bearer YOUR_API_KEY") req.Header.Set("Content-Type", "application/json") resp, _ := http.DefaultClient.Do(req) defer resp.Body.Close() data, _ := io.ReadAll(resp.Body) // 响应体中 message.reasoning_content 为思考过程，message.content 为最终回答 fmt.Println(string(data))}

六、调用示例：关闭深度思考

cURLPythonNode.jsJavaGocurl -X POST 'https://tokenhub.tencentmaas.com/v1/chat/completions' \\ -H 'Content-Type: application/json' \\ -H 'Authorization: Bearer YOUR_API_KEY' \\ -d '{ "model": "MODEL 参数值", "messages": [ {"role": "user", "content": "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？"} ], "thinking": {"type": "disabled"}, "stream": false }'from openai import OpenAIclient = OpenAI( api_key="YOUR_API_KEY", base_url="https://tokenhub.tencentmaas.com/v1",)response = client.chat.completions.create( model="MODEL 参数值", messages=[ {"role": "user", "content": "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？"}, ], extra_body={"thinking": {"type": "disabled"}},)print(response.choices[0].message.content)import OpenAI from 'openai';const client = new OpenAI({ apiKey: 'YOUR_API_KEY', baseURL: 'https://tokenhub.tencentmaas.com/v1',});const response = await client.chat.completions.create({ model: 'MODEL 参数值', messages: [ { role: 'user', content: '小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？' }, ], thinking: { type: 'disabled' },} as any);console.log(response.choices[0].message.content);import okhttp3.*;import com.google.gson.Gson;import java.util.*;public class DisableThinking { public static void main(String[] args) throws Exception { Map<String, Object> body = new HashMap<>(); body.put("model", "MODEL 参数值"); body.put("messages", List.of( Map.of("role", "user", "content", "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？") )); body.put("thinking", Map.of("type", "disabled")); Request request = new Request.Builder() .url("https://tokenhub.tencentmaas.com/v1/chat/completions") .header("Authorization", "Bearer YOUR_API_KEY") .post(RequestBody.create(new Gson().toJson(body), MediaType.parse("application/json"))) .build(); try (Response response = new OkHttpClient().newCall(request).execute()) { System.out.println(response.body().string()); } }}package mainimport ( "bytes" "encoding/json" "fmt" "io" "net/http")func main() { body, _ := json.Marshal(map[string]interface{}{ "model": "MODEL 参数值", "messages": []map[string]string{ {"role": "user", "content": "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？"}, }, "thinking": map[string]string{"type": "disabled"}, }) req, _ := http.NewRequest("POST", "https://tokenhub.tencentmaas.com/v1/chat/completions", bytes.NewBuffer(body)) req.Header.Set("Authorization", "Bearer YOUR_API_KEY") req.Header.Set("Content-Type", "application/json") resp, _ := http.DefaultClient.Do(req) defer resp.Body.Close() data, _ := io.ReadAll(resp.Body) fmt.Println(string(data))}

七、推理深度配置

通过 reasoning_effort 参数控制推理深度。该参数用于约束模型投入多少推理强度；推理强度越高，通常回答会更充分，但延迟和 token 消耗也会更高。

八、支持模型

九、调用示例：推理深度配置

cURLPythonNode.jsJavaGocurl -X POST 'https://tokenhub.tencentmaas.com/v1/chat/completions' \\ -H 'Content-Type: application/json' \\ -H 'Authorization: Bearer YOUR_API_KEY' \\ -d '{ "model": "MODEL 参数值", "messages": [ {"role": "user", "content": "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？"} ], "stream": false, "temperature": 0.9, "reasoning_effort": "high" }'from openai import OpenAIclient = OpenAI( api_key="YOUR_API_KEY", base_url="https://tokenhub.tencentmaas.com/v1",)response = client.chat.completions.create( model="MODEL 参数值", messages=[ {"role": "user", "content": "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？"}, ], temperature=0.9, extra_body={"reasoning_effort": "high"},)msg = response.choices[0].messageif hasattr(msg, "reasoning_content"): print("思考过程:", getattr(msg, "reasoning_content"))print("最终回答:", msg.content)import OpenAI from 'openai';const client = new OpenAI({ apiKey: 'YOUR_API_KEY', baseURL: 'https://tokenhub.tencentmaas.com/v1',});const response = await client.chat.completions.create({ model: 'MODEL 参数值', messages: [ { role: 'user', content: '小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？' }, ], temperature: 0.9, reasoning_effort: 'high',} as any);const msg: any = response.choices[0].message;if (msg.reasoning_content) console.log('思考过程:', msg.reasoning_content);console.log('最终回答:', msg.content);import okhttp3.*;import com.google.gson.Gson;import java.util.*;public class ReasoningEffortChat { public static void main(String[] args) throws Exception { Map<String, Object> body = new HashMap<>(); body.put("model", "MODEL 参数值"); body.put("messages", List.of( Map.of("role", "user", "content", "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？") )); body.put("temperature", 0.9); body.put("reasoning_effort", "high"); Request request = new Request.Builder() .url("https://tokenhub.tencentmaas.com/v1/chat/completions") .header("Authorization", "Bearer YOUR_API_KEY") .post(RequestBody.create(new Gson().toJson(body), MediaType.parse("application/json"))) .build(); try (Response response = new OkHttpClient().newCall(request).execute()) { System.out.println(response.body().string()); } }}package mainimport ( "bytes" "encoding/json" "fmt" "io" "net/http")func main() { body, _ := json.Marshal(map[string]interface{}{ "model": "MODEL 参数值", "messages": []map[string]string{ {"role": "user", "content": "小明有5个苹果，给了小红2个，又买了3个，最后还剩几个？"}, }, "temperature": 0.9, "reasoning_effort": "high", }) req, _ := http.NewRequest("POST", "https://tokenhub.tencentmaas.com/v1/chat/completions", bytes.NewBuffer(body)) req.Header.Set("Authorization", "Bearer YOUR_API_KEY") req.Header.Set("Content-Type", "application/json") resp, _ := http.DefaultClient.Do(req) defer resp.Body.Close() data, _ := io.ReadAll(resp.Body) fmt.Println(string(data))}

十、响应示例

启用思考后，响应中会附带 reasoning_content 思考过程字段：

{ "id": "REPLACED_ID", "object": "chat.completion", "created": 1775146546, "model": "MODEL 参数值", "choices": [ { "index": 0, "message": { "role": "assistant", "content": "答案是 6 个。", "reasoning_content": "用户现在需要解决的是小明苹果数量变化的问题，首先得理清楚每一步的变化。首先小明一开始有5个苹果，给了小红2个，那这时候应该减去2，对吧？然后又买了3个，这时候要加上3。所以计算的话就是5减2再加3。先算5-2=3，然后3+3=6。" }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": 22, "completion_tokens": 264, "total_tokens": 286 }}

思考模式下的工具调用，需在每一轮请求都回填历史 reasoning_content，以获取最佳效果，详情请参见 保留式思考模式（Preserved Thinking）。