一、概述
通用调用方式（BaseURL、API Key、messages 拼接、Function Calling、流式输出等）请参见 语言模型调用概览。这篇仅介绍 GLM 系列的特有能力和使用方式。
二、前提条件
·已注册腾讯云账号并开通 TokenHub 服务。
·已在 TokenHub 控制台 获取 API Key。
·已根据所用语言安装对应 SDK 或具备 HTTP 请求能力。
三、快速开始
以下示例展示如何调用 glm-5.3 模型完成一次基础对话。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 'Authorization: Bearer YOUR_API_KEY' \\ -H 'Content-Type: application/json' \\ -d '{ "model": "glm-5.3", "messages": [{"role": "user", "content": "你好"}], "max_tokens": 1024 }'from openai import OpenAIclient = OpenAI( api_key="YOUR_API_KEY", base_url="https://tokenhub.tencentmaas.com/v1",)response = client.chat.completions.create( model="glm-5.3", messages=[{"role": "user", "content": "你好"}], max_tokens=1024,)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: 'glm-5.3', messages: [{ role: 'user', content: '你好' }], max_tokens: 1024,});console.log(response.choices[0].message.content);import okhttp3.*;import com.google.gson.Gson;import java.util.*;public class GlmQuickStart { public static void main(String[] args) throws Exception { Map<String, Object> body = new HashMap<>(); body.put("model", "glm-5.3"); body.put("messages", List.of(Map.of("role", "user", "content", "你好"))); body.put("max_tokens", 1024); 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": "glm-5.3", "messages": []map[string]string{{"role": "user", "content": "你好"}}, "max_tokens": 1024, }) 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))}
注意：说明：
注意：GLM 系列模型默认开启思考模式，响应中会包含 reasoning_content 字段。如不需要思考过程，请参见下文 思考模式 章节关闭。
四、支持的模型
| model 参数值 | 定位 | 上下文 / 最大输出 | 多模态 | 推荐场景 |
| glm-5.3 | 旗舰（最新） | 1M tokens / 128K tokens | 不支持 | 深度推理、复杂任务、Agent 与编码、长程任务 |
| glm-5.3-flashx | 轻量多模态（增强） | 1M tokens / 128K tokens | 支持图片、视频、文件 | 多模态理解、高速推理（200 tokens/s）、日常对话与创作 |
| glm-5.3-flash | 轻量多模态 | 1M tokens / 128K tokens | 支持图片、视频、文件 | 多模态理解、快速推理、日常对话与创作 |
| glm-5.2 | 旗舰 | 1M tokens / 128K tokens | 不支持 | Agent 与编码、长程任务、全栈开发、代码翻译、科研复现 |
| glm-5v-turbo（即将下线） | 多模态 | 200K tokens / 128K tokens | 支持图片、视频、文件 | 图像理解、视频分析、文档解析 |
五、思考模式
GLM 系列将对话能力与推理能力合为一体，可通过 thinking 参数控制是否启用思考能力，无需切换 model 参数。
六、开启或关闭思考
通过 thinking 字段控制思考行为。该字段为对象格式，包含一个 type 属性：
| 字段 | 类型 | 取值 | 默认值 | 说明 |
| type | String | enabled / disabled | enabled | 控制当前请求是否启用思考能力。GLM-5.3、GLM-5.3-Flash 和 GLM-5.3-FlashX 默认启用且不支持关闭（即不支持 disabled 值），传入 disabled 会导致请求失败 |
以下示例展示如何在请求中关闭思考（以 glm-5.2 为例；glm-5.3、glm-5.3-flash 和 glm-5.3-flashx 不支持关闭思考，传入 disabled 将导致请求失败）：
cURLPythonNode.jsJavaGocurl -X POST 'https://tokenhub.tencentmaas.com/v1/chat/completions' \\ -H 'Authorization: Bearer YOUR_API_KEY' \\ -H 'Content-Type: application/json' \\ -d '{ "model": "glm-5.2", "messages": [{"role": "user", "content": "你好"}], "thinking": {"type": "disabled"} }'response = client.chat.completions.create( model="glm-5.2", messages=[{"role": "user", "content": "你好"}], extra_body={"thinking": {"type": "disabled"}},)const response = await client.chat.completions.create({ model: 'glm-5.2', messages: [{ role: 'user', content: '你好' }], // @ts-ignore – thinking 为 GLM 扩展字段 thinking: { type: 'disabled' },});import okhttp3.*;import com.google.gson.Gson;import java.util.*;public class GlmThinkingDisabled { public static void main(String[] args) throws Exception { Map<String, Object> body = new HashMap<>(); body.put("model", "glm-5.2"); body.put("messages", List.of(Map.of("role", "user", "content", "你好"))); 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": "glm-5.2", "messages": []map[string]string{{"role": "user", "content": "你好"}}, "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))}
注意：说明：
注意：thinking 不是 OpenAI 标准字段。使用 OpenAI SDK 时需通过 SDK 提供的额外字段机制（Python 用 extra_body、Node.js 直接传字段）；HTTP 直接调用时放在请求体顶层。
七、获取思考内容
开启思考后，响应中会新增 reasoning_content 字段，与 content 同级：
{ "choices": [{ "message": { "role": "assistant", "reasoning_content": "让我分析一下这个问题…", "content": "最终答案是…" } }]}
由于 reasoning_content 不是 OpenAI 标准字段，使用各语言 SDK 时需通过判空或反射方式访问；HTTP 直接调用时直接读取响应 JSON 即可。
PythonNode.jsJavaGomessage = response.choices[0].messageif hasattr(message, "reasoning_content") and message.reasoning_content: print("思考过程：", message.reasoning_content)print("回答：", message.content)const message = response.choices[0].message;// @ts-ignore – reasoning_content 为 GLM 扩展字段if (message.reasoning_content) { console.log('思考过程：', message.reasoning_content);}console.log('回答：', message.content);// HTTP 调用拿到响应字符串后，用 Gson 解析 reasoning_content 和 content 字段import com.google.gson.JsonObject;import com.google.gson.JsonParser;String respBody = response.body().string();JsonObject json = JsonParser.parseString(respBody).getAsJsonObject();JsonObject message = json.getAsJsonArray("choices") .get(0).getAsJsonObject() .getAsJsonObject("message");if (message.has("reasoning_content") && !message.get("reasoning_content").isJsonNull()) { System.out.println("思考过程：" + message.get("reasoning_content").getAsString());}System.out.println("回答：" + message.get("content").getAsString());type Message struct { Role string `json:"role"` Content string `json:"content"` ReasoningContent string `json:"reasoning_content,omitempty"`}type Choice struct { Index int `json:"index"` Message Message `json:"message"`}type ChatResponse struct { Choices []Choice `json:"choices"`}var result ChatResponsejson.Unmarshal(data, &result)if result.Choices[0].Message.ReasoningContent != "" { fmt.Println("思考过程：", result.Choices[0].Message.ReasoningContent)}fmt.Println("回答：", result.Choices[0].Message.Content)
八、多轮对话处理
构建后续轮次的 messages 时，无需回写 reasoning_content。仅将 content 字段作为 assistant 消息传入即可。
九、流式调用
启用思考模式时建议使用流式调用（stream=True）。思考内容可能较长，非流式调用容易触发网关超时。
流式模式下，reasoning_content 会在 content 之前完整输出。客户端处理逻辑：累积 delta.reasoning_content 输出思考过程，再累积 delta.content 输出最终回答。
cURLPythonNode.jsJavaGocurl -N -X POST 'https://tokenhub.tencentmaas.com/v1/chat/completions' \\ -H 'Authorization: Bearer YOUR_API_KEY' \\ -H 'Content-Type: application/json' \\ -d '{ "model": "glm-5.3", "messages": [{"role": "user", "content": "解释量子纠缠"}], "stream": true, "stream_options": {"include_usage": true}, "thinking": {"type": "enabled"} }'stream = client.chat.completions.create( model="glm-5.3", messages=[{"role": "user", "content": "解释量子纠缠"}], stream=True, stream_options={"include_usage": True}, extra_body={"thinking": {"type": "enabled"}},)is_answering = Falsefor chunk in stream: if not chunk.choices: continue delta = chunk.choices[0].delta if hasattr(delta, "reasoning_content") and delta.reasoning_content: print(delta.reasoning_content, end="", flush=True) if hasattr(delta, "content") and delta.content: if not is_answering: print("\\n— 回答 —\\n") is_answering = True print(delta.content, end="", flush=True)const stream = await client.chat.completions.create({ model: 'glm-5.3', messages: [{ role: 'user', content: '解释量子纠缠' }], stream: true, stream_options: { include_usage: true }, // @ts-ignore – thinking 为 GLM 扩展字段 thinking: { type: 'enabled' },});let isAnswering = false;for await (const chunk of stream) { if (!chunk.choices?.length) continue; const delta = chunk.choices[0].delta; // @ts-ignore – reasoning_content 为 GLM 扩展字段 if (delta.reasoning_content) { process.stdout.write(delta.reasoning_content); } if (delta.content) { if (!isAnswering) { process.stdout.write('\\n— 回答 —\\n'); isAnswering = true; } process.stdout.write(delta.content); }}import okhttp3.*;import com.google.gson.*;import java.util.*;import java.io.BufferedReader;import java.io.InputStreamReader;public class GlmStream { public static void main(String[] args) throws Exception { Map<String, Object> body = new HashMap<>(); body.put("model", "glm-5.3"); body.put("messages", List.of(Map.of("role", "user", "content", "解释量子纠缠"))); body.put("stream", true); body.put("stream_options", Map.of("include_usage", true)); 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(); BufferedReader reader = new BufferedReader( new InputStreamReader(response.body().byteStream()))) { String line; boolean isAnswering = false; while ((line = reader.readLine()) != null) { if (!line.startsWith("data: ")) continue; String data = line.substring(6); if (data.equals("[DONE]")) break; JsonObject chunk = JsonParser.parseString(data).getAsJsonObject(); JsonArray choices = chunk.getAsJsonArray("choices"); if (choices == null || choices.size() == 0) continue; JsonObject delta = choices.get(0).getAsJsonObject().getAsJsonObject("delta"); if (delta.has("reasoning_content") && !delta.get("reasoning_content").isJsonNull()) { System.out.print(delta.get("reasoning_content").getAsString()); } if (delta.has("content") && !delta.get("content").isJsonNull()) { if (!isAnswering) { System.out.println("\\n— 回答 —"); isAnswering = true; } System.out.print(delta.get("content").getAsString()); } } } }}package mainimport ( "bufio" "bytes" "encoding/json" "fmt" "net/http" "strings")type StreamDelta struct { Content string `json:"content,omitempty"` ReasoningContent string `json:"reasoning_content,omitempty"`}type StreamChoice struct { Delta StreamDelta `json:"delta"`}type StreamChunk struct { Choices []StreamChoice `json:"choices"`}func main() { body, _ := json.Marshal(map[string]interface{}{ "model": "glm-5.3", "messages": []map[string]string{{"role": "user", "content": "解释量子纠缠"}}, "stream": true, "stream_options": map[string]bool{"include_usage": true}, "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() isAnswering := false scanner := bufio.NewScanner(resp.Body) for scanner.Scan() { line := scanner.Text() if !strings.HasPrefix(line, "data: ") { continue } data := strings.TrimPrefix(line, "data: ") if data == "[DONE]" { break } var chunk StreamChunk if err := json.Unmarshal([]byte(data), &chunk); err != nil { continue } if len(chunk.Choices) == 0 { continue } delta := chunk.Choices[0].Delta if delta.ReasoningContent != "" { fmt.Print(delta.ReasoningContent) } if delta.Content != "" { if !isAnswering { fmt.Println("\\n— 回答 —") isAnswering = true } fmt.Print(delta.Content) } }}
十、思考深度（reasoning_effort）
glm-5.2、glm-5.3、glm-5.3-flash 和 glm-5.3-flashx 在 thinking 字段之外，额外提供 reasoning_effort 参数用于控制思考深度，便于在效果与延迟之间灵活权衡。该参数仅 glm-5.2、glm-5.3、glm-5.3-flash 和 glm-5.3-flashx 支持，其他 GLM 模型请使用 thinking 字段控制（参见上文 开启或关闭思考）。不传时默认 max。
| 模型 | 取值 | 说明 |
| glm-5.3 / glm-5.3-flash / glm-5.3-flashx | max | 深度推理（默认值）。适合复杂推理、长程任务、深度代码分析等场景。 |
| | high | 增强推理。适合常规推理、代码生成等场景。 |
| | low | 轻度推理。适合简单问答、快速响应等对延迟敏感的场景。 |
| glm-5.2 | max、xhigh | 深度推理（默认值，xhigh 效果等同 max）。适合复杂推理、长程任务等场景。 |
| | high、medium、low | 增强推理（三者效果等同）。适合常规推理、代码生成等场景。 |
| | minimal、none | 放弃思考，等同于关闭思考。 |
reasoning_effort 不是 OpenAI 标准字段。使用 OpenAI SDK 时需通过 SDK 提供的额外字段机制（Python 用 extra_body、Node.js 直接传字段）；HTTP 直接调用时放在请求体顶层。示例如下：
cURLPythonNode.jscurl –location -X POST 'https://tokenhub.tencentmaas.com/v1/chat/completions' \\ –header "Authorization: Bearer ${API_KEY}" \\ –header 'Content-Type: application/json' \\ –data-raw '{ "model": "glm-5.2", "messages": [ {"role": "user", "content": "请分析以下代码的性能瓶颈并给出优化方案。"} ], "thinking": {"type": "enabled"}, "reasoning_effort": "max", "temperature": 1.0, "max_tokens": 65536 }'from openai import OpenAIclient = OpenAI( api_key="YOUR_API_KEY", base_url="https://tokenhub.tencentmaas.com/v1")response = client.chat.completions.create( model="glm-5.2", messages=[ {"role": "user", "content": "请证明根号 2 是无理数。"} ], extra_body={ "thinking": {"type": "enabled"}, "reasoning_effort": "max", }, temperature=1.0, max_tokens=65536,)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 completion = await client.chat.completions.create({ model: 'glm-5.2', messages: [ { role: 'user', content: '请分析这段代码的复杂度并给出优化建议。' }, ], // @ts-ignore – thinking / reasoning_effort 为 GLM 扩展字段 thinking: { type: 'enabled' }, // @ts-ignore reasoning_effort: 'max', temperature: 1.0, max_tokens: 65536,});console.log(completion.choices[0].message.content);
注意：说明：
十一、推荐参数
为获得最佳效果，建议使用以下参数配置：
| 参数 | 推荐值 | 说明 |
| temperature | 1（默认值） | GLM-5.3、GLM-5.3-Flash 和 GLM-5.3-FlashX 推荐使用默认值，可获得更稳定的输出质量。 |
| top_p | 0.95（默认值） | 与 temperature 配合使用，控制采样范围。 |
注意：说明：
注意：以上参数为 GLM-5.3、GLM-5.3-Flash 和 GLM-5.3-FlashX 的推荐配置。其他 GLM 版本可能需要根据实际情况调整。
十二、工具调用
GLM 系列的 Function Calling 遵循 OpenAI 标准协议（tools / tool_choice）。通用用法请参见 调用概览。
十三、流式场景下的参数拼接
GLM 模型在流式调用时，tool_call.arguments 会分多个 chunk 增量返回，客户端需要按 tool_call.index 累积拼接：
PythonNode.jsJavaGotools = [{ "type": "function", "function": { "name": "get_weather", "description": "查询指定城市的天气", "parameters": { "type": "object", "properties": { "city": {"type": "string", "description": "城市名称"} }, "required": ["city"] } }}]completion = client.chat.completions.create( model="glm-5.2", messages=[{"role": "user", "content": "查询深圳天气"}], tools=tools, stream=True, extra_body={"thinking": {"type": "disabled"}},)arg_buffer = ""tool_name = ""for chunk in completion: if not chunk.choices: continue delta = chunk.choices[0].delta if hasattr(delta, "tool_calls") and delta.tool_calls: for tc in delta.tool_calls: if tc.function and tc.function.name: tool_name = tc.function.name if tc.function and tc.function.arguments: arg_buffer += tc.function.argumentsprint(f"调用工具：{tool_name}")print(f"完整参数：{arg_buffer}") # 输出：{"city": "深圳"}const tools = [{ type: 'function', function: { name: 'get_weather', description: '查询指定城市的天气', parameters: { type: 'object', properties: { city: { type: 'string', description: '城市名称' } }, required: ['city'] } }}];const completion = await client.chat.completions.create({ model: 'glm-5.2', messages: [{ role: 'user', content: '查询深圳天气' }], tools: tools, stream: true, // @ts-ignore – thinking 为 GLM 扩展字段 thinking: { type: 'disabled' },});let argBuffer = '';let toolName = '';for await (const chunk of completion) { if (!chunk.choices?.length) continue; const delta = chunk.choices[0].delta; if (delta.tool_calls) { for (const tc of delta.tool_calls) { if (tc.function?.name) toolName = tc.function.name; if (tc.function?.arguments) argBuffer += tc.function.arguments; } }}console.log(`调用工具：${toolName}`);console.log(`完整参数：${argBuffer}`); // 输出：{"city": "深圳"}import okhttp3.*;import com.google.gson.*;import java.util.*;import java.io.BufferedReader;import java.io.InputStreamReader;List<Map<String, Object>> tools = List.of(Map.of( "type", "function", "function", Map.of( "name", "get_weather", "description", "查询指定城市的天气", "parameters", Map.of( "type", "object", "properties", Map.of("city", Map.of("type", "string", "description", "城市名称")), "required", List.of("city") ) )));Map<String, Object> body = new HashMap<>();body.put("model", "glm-5.2");body.put("messages", List.of(Map.of("role", "user", "content", "查询深圳天气")));body.put("tools", tools);body.put("stream", true);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();StringBuilder argBuffer = new StringBuilder();String toolName = "";try (Response response = new OkHttpClient().newCall(request).execute(); BufferedReader reader = new BufferedReader( new InputStreamReader(response.body().byteStream()))) { String line; while ((line = reader.readLine()) != null) { if (!line.startsWith("data: ")) continue; String data = line.substring(6); if (data.equals("[DONE]")) break; JsonObject chunk = JsonParser.parseString(data).getAsJsonObject(); JsonArray choices = chunk.getAsJsonArray("choices"); if (choices == null || choices.size() == 0) continue; JsonObject delta = choices.get(0).getAsJsonObject().getAsJsonObject("delta"); if (delta.has("tool_calls") && delta.get("tool_calls").isJsonArray()) { for (JsonElement tcEl : delta.getAsJsonArray("tool_calls")) { JsonObject fn = tcEl.getAsJsonObject().getAsJsonObject("function"); if (fn != null) { if (fn.has("name") && !fn.get("name").isJsonNull()) { toolName = fn.get("name").getAsString(); } if (fn.has("arguments") && !fn.get("arguments").isJsonNull()) { argBuffer.append(fn.get("arguments").getAsString()); } } } } }}System.out.println("调用工具：" + toolName);System.out.println("完整参数：" + argBuffer);package mainimport ( "bufio" "bytes" "encoding/json" "fmt" "net/http" "strings")type ToolCallFunction struct { Name string `json:"name,omitempty"` Arguments string `json:"arguments,omitempty"`}type ToolCall struct { Index int `json:"index"` Function ToolCallFunction `json:"function"`}type ToolStreamDelta struct { ToolCalls []ToolCall `json:"tool_calls,omitempty"`}type ToolStreamChoice struct { Delta ToolStreamDelta `json:"delta"`}type ToolStreamChunk struct { Choices []ToolStreamChoice `json:"choices"`}func main() { tools := []map[string]interface{}{{ "type": "function", "function": map[string]interface{}{ "name": "get_weather", "description": "查询指定城市的天气", "parameters": map[string]interface{}{ "type": "object", "properties": map[string]interface{}{ "city": map[string]interface{}{"type": "string", "description": "城市名称"}, }, "required": []string{"city"}, }, }, }} body, _ := json.Marshal(map[string]interface{}{ "model": "glm-5.2", "messages": []map[string]string{{"role": "user", "content": "查询深圳天气"}}, "tools": tools, "stream": true, "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() var argBuffer strings.Builder var toolName string scanner := bufio.NewScanner(resp.Body) for scanner.Scan() { line := scanner.Text() if !strings.HasPrefix(line, "data: ") { continue } data := strings.TrimPrefix(line, "data: ") if data == "[DONE]" { break } var chunk ToolStreamChunk if err := json.Unmarshal([]byte(data), &chunk); err != nil { continue } if len(chunk.Choices) == 0 { continue } for _, tc := range chunk.Choices[0].Delta.ToolCalls { if tc.Function.Name != "" { toolName = tc.Function.Name } if tc.Function.Arguments != "" { argBuffer.WriteString(tc.Function.Arguments) } } } fmt.Printf("调用工具：%s\\n完整参数：%s\\n", toolName, argBuffer.String())}
十四、tool_stream 参数
GLM 系列支持 tool_stream 参数（Boolean 类型），用于控制工具调用参数的流式分片粒度：
extra_body={"tool_stream": True, "thinking": {"type": "disabled"}} # thinking 关闭仅适用于 glm-5.2、glm-5v-turbo 等支持关闭的模型，glm-5.3 系列传入 disabled 会报错
十五、多模态调用
glm-5v-turbo、glm-5.3-flash 和 glm-5.3-flashx 支持 GLM 系列中的多模态输入，支持图像、视频和文件输入，输出为文本。
十六、使用限制
·图像、视频、文件不可在同一请求中混合传入。
·文件输入仅支持 URL，不支持 Base64。
·图像输入支持 URL 和 Base64 两种方式。
十七、图像输入
已验证支持的格式：PNG、JPG、JPEG、WebP。其他格式如需使用，请先进行小样本测试确认。
cURLPythonNode.jsJavaGocurl -X POST 'https://tokenhub.tencentmaas.com/v1/chat/completions' \\ -H 'Authorization: Bearer YOUR_API_KEY' \\ -H 'Content-Type: application/json' \\ -d '{ "model": "glm-5v-turbo", "messages": [{ "role": "user", "content": [ {"type": "text", "text": "请描述这张图片"}, {"type": "image_url", "image_url": {"url": "https://example.com/photo.png"}} ] }], "max_tokens": 1024, "thinking": {"type": "disabled"} }'response = client.chat.completions.create( model="glm-5v-turbo", messages=[{ "role": "user", "content": [ {"type": "text", "text": "请描述这张图片"}, {"type": "image_url", "image_url": {"url": "https://example.com/photo.png"}}, ], }], max_tokens=1024, extra_body={"thinking": {"type": "disabled"}},)print(response.choices[0].message.content)const response = await client.chat.completions.create({ model: 'glm-5v-turbo', messages: [{ role: 'user', content: [ { type: 'text', text: '请描述这张图片' }, { type: 'image_url', image_url: { url: 'https://example.com/photo.png' } }, ], }], max_tokens: 1024, // @ts-ignore – thinking 为 GLM 扩展字段 thinking: { type: 'disabled' },});console.log(response.choices[0].message.content);import okhttp3.*;import com.google.gson.Gson;import java.util.*;public class GlmImageInput { public static void main(String[] args) throws Exception { List<Map<String, Object>> content = List.of( Map.of("type", "text", "text", "请描述这张图片"), Map.of("type", "image_url", "image_url", Map.of("url", "https://example.com/photo.png")) ); Map<String, Object> body = new HashMap<>(); body.put("model", "glm-5v-turbo"); body.put("messages", List.of(Map.of("role", "user", "content", content))); body.put("max_tokens", 1024); 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": "glm-5v-turbo", "messages": []map[string]interface{}{ { "role": "user", "content": []map[string]interface{}{ {"type": "text", "text": "请描述这张图片"}, {"type": "image_url", "image_url": map[string]string{ "url": "https://example.com/photo.png", }}, }, }, }, "max_tokens": 1024, "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))}
Base64 方式（Python 示例）：
将本地图片读取为 Base64 后，通过 Data URI 形式传入 image_url.url：
import base64with open("local.jpg", "rb") as f: b64 = base64.b64encode(f.read()).decode()response = client.chat.completions.create( model="glm-5v-turbo", messages=[{ "role": "user", "content": [ {"type": "text", "text": "图中有什么？"}, {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{b64}"}}, ], }], max_tokens=1024, extra_body={"thinking": {"type": "disabled"}},)
十八、视频输入
已验证支持的格式：MP4、MPEG、MOV、AVI、WebM、WMV、3GPP。其他格式如需使用，请先进行小样本测试确认。
将上文图像示例中的 image_url 字段替换为 video_url，即可传入视频文件 URL：
cURLPythonNode.jsJavaGocurl -X POST 'https://tokenhub.tencentmaas.com/v1/chat/completions' \\ -H 'Authorization: Bearer YOUR_API_KEY' \\ -H 'Content-Type: application/json' \\ -d '{ "model": "glm-5v-turbo", "messages": [{ "role": "user", "content": [ {"type": "text", "text": "请总结这段视频的内容"}, {"type": "video_url", "video_url": {"url": "https://example.com/demo.mp4"}} ] }], "max_tokens": 2048, "thinking": {"type": "disabled"} }'response = client.chat.completions.create( model="glm-5v-turbo", messages=[{ "role": "user", "content": [ {"type": "text", "text": "请总结这段视频的内容"}, {"type": "video_url", "video_url": {"url": "https://example.com/demo.mp4"}}, ], }], max_tokens=2048, extra_body={"thinking": {"type": "disabled"}},)print(response.choices[0].message.content)const response = await client.chat.completions.create({ model: 'glm-5v-turbo', messages: [{ role: 'user', content: [ { type: 'text', text: '请总结这段视频的内容' }, { type: 'video_url', video_url: { url: 'https://example.com/demo.mp4' } }, ], }], max_tokens: 2048, // @ts-ignore – thinking 为 GLM 扩展字段 thinking: { type: 'disabled' },});console.log(response.choices[0].message.content);List<Map<String, Object>> content = List.of( Map.of("type", "text", "text", "请总结这段视频的内容"), Map.of("type", "video_url", "video_url", Map.of("url", "https://example.com/demo.mp4")));Map<String, Object> body = new HashMap<>();body.put("model", "glm-5v-turbo");body.put("messages", List.of(Map.of("role", "user", "content", content)));body.put("max_tokens", 2048);body.put("thinking", Map.of("type", "disabled"));// 其余 HTTP 请求逻辑同图像示例body, _ := json.Marshal(map[string]interface{}{ "model": "glm-5v-turbo", "messages": []map[string]interface{}{ { "role": "user", "content": []map[string]interface{}{ {"type": "text", "text": "请总结这段视频的内容"}, {"type": "video_url", "video_url": map[string]string{ "url": "https://example.com/demo.mp4", }}, }, }, }, "max_tokens": 2048, "thinking": map[string]string{"type": "disabled"},})// 其余 HTTP 请求逻辑同图像示例
十九、文件输入
已验证支持的格式：PDF、TXT、DOC。其他格式如需使用，请先进行小样本测试确认。
文件输入仅支持通过 URL 传入，不支持 Base64 编码。如本地有文件需要解析，请先上传至对象存储服务（如腾讯云 COS），再使用生成的 URL。
cURLPythonNode.jsJavaGocurl -X POST 'https://tokenhub.tencentmaas.com/v1/chat/completions' \\ -H 'Authorization: Bearer YOUR_API_KEY' \\ -H 'Content-Type: application/json' \\ -d '{ "model": "glm-5v-turbo", "messages": [{ "role": "user", "content": [ {"type": "text", "text": "请提取这份文档的核心要点"}, {"type": "file_url", "file_url": {"url": "https://example.com/report.pdf"}} ] }], "max_tokens": 4096, "thinking": {"type": "disabled"} }'response = client.chat.completions.create( model="glm-5v-turbo", messages=[{ "role": "user", "content": [ {"type": "text", "text": "请提取这份文档的核心要点"}, {"type": "file_url", "file_url": {"url": "https://example.com/report.pdf"}}, ], }], max_tokens=4096, extra_body={"thinking": {"type": "disabled"}},)print(response.choices[0].message.content)const response = await client.chat.completions.create({ model: 'glm-5v-turbo', messages: [{ role: 'user', content: [ { type: 'text', text: '请提取这份文档的核心要点' }, { type: 'file_url', file_url: { url: 'https://example.com/report.pdf' } }, ], }], max_tokens: 4096, // @ts-ignore – thinking 为 GLM 扩展字段 thinking: { type: 'disabled' },});console.log(response.choices[0].message.content);List<Map<String, Object>> content = List.of( Map.of("type", "text", "text", "请提取这份文档的核心要点"), Map.of("type", "file_url", "file_url", Map.of("url", "https://example.com/report.pdf")));Map<String, Object> body = new HashMap<>();body.put("model", "glm-5v-turbo");body.put("messages", List.of(Map.of("role", "user", "content", content)));body.put("max_tokens", 4096);body.put("thinking", Map.of("type", "disabled"));// 其余 HTTP 请求逻辑同图像示例body, _ := json.Marshal(map[string]interface{}{ "model": "glm-5v-turbo", "messages": []map[string]interface{}{ { "role": "user", "content": []map[string]interface{}{ {"type": "text", "text": "请提取这份文档的核心要点"}, {"type": "file_url", "file_url": map[string]string{ "url": "https://example.com/report.pdf", }}, }, }, }, "max_tokens": 4096, "thinking": map[string]string{"type": "disabled"},})// 其余 HTTP 请求逻辑同图像示例
20、使用限制
| 限制项 | 说明 |
| 思考模式默认开启 | 不传 thinking 参数时默认启用思考，思考内容通过 reasoning_content 字段返回。glm-5.3、glm-5.3-flash 和 glm-5.3-flashx 不支持关闭思考；glm-5.2、glm-5v-turbo 等模型不需要思考时，可显式传入 thinking: {"type": "disabled"} 关闭。 |
| 非流式调用超时风险 | 思考模式下输出较长，建议使用 stream=True。 |
| 多模态仅限 glm-5v-turbo、glm-5.3-flash 和 glm-5.3-flashx | 其余模型不支持图像、视频、文件输入。 |
| 多模态输入不可混合 | 图像、视频、文件在同一请求中只能传入一类。 |
| 文件输入仅支持 URL | file_url 不支持 Base64 或 Data URI。 |
| 工具参数增量返回 | 流式调用时 tool_call.arguments 分多个 chunk 返回，需客户端拼接。 |
| 请求体积上限 | 单次请求 body 不超过 100 MB。 |
21、相关文档
·语言模型调用概览：TokenHub 语言模型通用调用文档，包含 BaseURL、API Key、多轮对话、Function Calling、Anthropic 协议等通用说明。
·TokenHub 控制台：API Key 创建与管理入口。