Python调用国产大模型API实战从DeepSeek到智谱GLM导语2026年4月国产大模型迎来爆发期。DeepSeek V4、美团LongCat-2.0、智谱GLM-5.1同日发布。本文手把手教你用Python调用这些国产大模型API成本只有OpenAI的1/5。一、为什么选国产大模型API国产大模型API的核心优势价格优势约$0.002/千Tokens是OpenAI的1/5中文理解天然优势本地化场景更精准响应速度国内服务器延迟更低合规可控数据不出境满足企业合规要求二、环境准备# 创建虚拟环境 python3 -m venv venv source venv/bin/activate # macOS/Linux # 或 venv\Scripts\activate # Windows # 安装依赖 pip install openai zhipuai requests三、DeepSeek API 调用DeepSeek是目前最热门的国产开源大模型支持1M上下文。import os from openai import OpenAI # DeepSeek API配置 client OpenAI( api_keyos.getenv(DEEPSEEK_API_KEY), base_urlhttps://api.deepseek.com/v1 ) def chat_deepseek(prompt, modeldeepseek-chat): 调用DeepSeek API response client.chat.completions.create( modelmodel, messages[ {role: system, content: 你是一个专业的Python编程助手。}, {role: user, content: prompt} ], temperature0.7, max_tokens2000 ) return response.choices[0].message.content # 示例让DeepSeek写一个快速排序 code_prompt 用Python实现快速排序算法要求 1. 使用原地排序in-place 2. 包含详细注释 3. 包含测试用例 result chat_deepseek(code_prompt) print(result)四、智谱GLM API 调用智谱GLM-5.1已能独立工作8小时支持Agent能力。import os from zhipuai import ZhipuAI # 智谱API配置 client ZhipuAI(api_keyos.getenv(ZHIPU_API_KEY)) def chat_glm(prompt, streamFalse): 调用智谱GLM API response client.chat.completions.create( modelglm-4-plus, # 或 glm-4, glm-4-flash 等 messages[ {role: user, content: prompt} ], streamstream, temperature0.8, top_p0.95 ) if stream: # 流式输出 for event in response: print(event.choices[0].delta.content, end, flushTrue) else: return response.choices[0].message.content # 示例分析一段代码 analysis_prompt 分析以下Python代码的性能瓶颈并给出优化建议 def slow_function(data): result [] for item in data: if item % 2 0: result.append(item * 2) return result result chat_glm(analysis_prompt) print(result)五、批量处理实战案例用国产大模型批量处理Excel数据。import pandas as pd from openai import OpenAI from concurrent.futures import ThreadPoolExecutor import os client OpenAI( api_keyos.getenv(DEEPSEEK_API_KEY), base_urlhttps://api.deepseek.com/v1 ) def analyze_sentiment(text, client): 分析文本情感 response client.chat.completions.create( modeldeepseek-chat, messages[ {role: system, content: 你是情感分析专家只返回Positive、Neutral或Negative。}, {role: user, content: f分析以下文本情感{text}} ], max_tokens10 ) return response.choices[0].message.content.strip() def batch_analyze(df, text_column, max_workers5): 批量处理DataFrame texts df[text_column].tolist() results [] # 使用线程池并发调用 with ThreadPoolExecutor(max_workersmax_workers) as executor: futures [executor.submit(analyze_sentiment, text, client) for text in texts] results [f.result() for f in futures] return results # 使用示例 if __name__ __main__: # 读取数据 df pd.read_excel(comments.xlsx) # 批量分析 sentiments batch_analyze(df, comment_text, max_workers10) # 保存结果 df[sentiment] sentiments df.to_excel(comments_analyzed.xlsx, indexFalse) print(分析完成)六、成本对比模型价格($/千Tokens)优势GPT-5.5$0.01综合能力强DeepSeek V4$0.002性价比高智谱GLM-4$0.001中文理解强美团LongCat$0.002本地生活场景七、注意事项API Key安全不要硬编码使用环境变量错误处理添加重试机制防止调用失败并发控制遵守API调用频率限制成本监控设置预算告警避免意外账单八、完整项目代码 国产大模型API调用封装 支持DeepSeek、智谱GLM、美团LongCat import os from typing import Optional, List, Dict from dataclasses import dataclass from enum import Enum class ModelType(Enum): DEEPSEEK deepseek ZHIPU zhipu MEITUAN meituan dataclass class ModelConfig: api_key: str base_url: Optional[str] None model: str deepseek-chat class ChineseLLMClient: def __init__(self, model_type: ModelType, api_key: str None): self.model_type model_type self.config self._get_config(model_type, api_key) self.client self._init_client() def _get_config(self, model_type: ModelType, api_key: str None) - ModelConfig: configs { ModelType.DEEPSEEK: ModelConfig( api_keyapi_key or os.getenv(DEEPSEEK_API_KEY), base_urlhttps://api.deepseek.com/v1, modeldeepseek-chat ), ModelType.ZHIPU: ModelConfig( api_keyapi_key or os.getenv(ZHIPU_API_KEY), base_urlhttps://open.bigmodel.cn/api/paas/v4, modelglm-4-plus ), ModelType.MEITUAN: ModelConfig( api_keyapi_key or os.getenv(MEITUAN_API_KEY), base_urlhttps://api.meituan.com/llm/v1, modellongcat-2.0 ) } return configs[model_type] def _init_client(self): from openai import OpenAI return OpenAI( api_keyself.config.api_key, base_urlself.config.base_url ) def chat(self, prompt: str, system: str None) - str: messages [] if system: messages.append({role: system, content: system}) messages.append({role: user, content: prompt}) response self.client.chat.completions.create( modelself.config.model, messagesmessages, temperature0.7, max_tokens2000 ) return response.choices[0].message.content # 使用示例 if __name__ __main__: # 使用DeepSeek client ChineseLLMClient(ModelType.DEEPSEEK) result client.chat(用Python写一个Hello World) print(result)结语国产大模型API已经非常成熟价格只有OpenAI的1/5。对于国内项目强烈推荐使用国产大模型。完整代码已上传到GitHubhttps://github.com/example/chinese-llm-api标签Python, DeepSeek, 智谱GLM, 大模型API, AI开发