OpenAI API vs Anthropic API vs Google Gemini API：2026年开发者指南

选择合适的LLM API提供商会影响整个应用架构。以下是从开发者角度出发的对比，包含真实代码示例。

环境配置

python
OpenAI
from openai import OpenAI
openai_client = OpenAI(api_key=os.environ['OPENAI_API_KEY'])
Anthropic
import anthropic
anthropic_client = anthropic.Anthropic(api_key=os.environ['ANTHROPIC_API_KEY'])
Google Gemini
import google.generativeai as genai
genai.configure(api_key=os.environ['GOOGLE_API_KEY'])
gemini_model = genai.GenerativeModel('gemini-2.5-pro')

流式响应

python
OpenAI 流式
with openai_client.chat.completions.stream(
    model='gpt-5',
    messages=[{'role': 'user', 'content': '解释机器学习中的Transformer'}]
) as stream:
    for text in stream.text_stream:
        print(text, end='', flush=True)
Anthropic 流式
with anthropic_client.messages.stream(
    model='claude-sonnet-4-5',
    max_tokens=2000,
    messages=[{'role': 'user', 'content': '解释机器学习中的Transformer'}]
) as stream:
    for text in stream.text_stream:
        print(text, end='', flush=True)
Gemini 流式
for chunk in gemini_model.generate_content('解释机器学习中的Transformer', stream=True):
    print(chunk.text, end='', flush=True)

函数调用

python
OpenAI 函数调用
tools = [{'type': 'function', 'function': {
    'name': 'get_weather',
    'description': '获取当前天气',
    'parameters': {'type': 'object', 'properties': {'city': {'type': 'string'}}, 'required': ['city']}
}}]
response = openai_client.chat.completions.create(
    model='gpt-5',
    messages=[{'role': 'user', 'content': '东京的天气？'}],
    tools=tools, tool_choice='auto'
)
if response.choices[0].message.tool_calls:
    tc = response.choices[0].message.tool_calls[0]
    print(f'调用: {tc.function.name}({tc.function.arguments})')
Anthropic 工具使用
tools_claude = [{'name': 'get_weather', 'description': '获取当前天气',
    'input_schema': {'type': 'object', 'properties': {'city': {'type': 'string'}}, 'required': ['city']}}]
response = anthropic_client.messages.create(
    model='claude-sonnet-4-5', max_tokens=1000,
    tools=tools_claude,
    messages=[{'role': 'user', 'content': '东京的天气？'}]
)
for block in response.content:
    if block.type == 'tool_use':
        print(f'工具: {block.name}, 输入: {block.input}')

结构化输出

python
from pydantic import BaseModel
from typing import List
class TechStack(BaseModel):
    languages: List[str]
    frameworks: List[str]
    databases: List[str]
OpenAI - 原生结构化输出（最可靠）
response = openai_client.beta.chat.completions.parse(
    model='gpt-5',
    messages=[{'role': 'user', 'content': '提取技术栈：Python/Django/PostgreSQL/AWS'}],
    response_format=TechStack
)
stack = response.choices[0].message.parsed
print(stack.languages)  # ['Python']

速率限制与重试

python
import time
import random
def with_retry(func, max_retries: int = 5):
    for attempt in range(max_retries):
        try:
            return func()
        except (openai.RateLimitError, anthropic.RateLimitError):
            if attempt == max_retries - 1:
                raise
            wait = (2 ** attempt) + random.uniform(0, 1)
            time.sleep(wait)

2026年成本对比

模型输入/百万Token输出/百万Token上下文长度

GPT-5$10$30128K GPT-5 mini$0.40$1.60128K Claude Sonnet 4$3$15200K Claude Haiku 4$0.80$4200K Gemini 2.5 Pro$1.25$101M Gemini Flash$0.075$0.301M

提供商选择指南

OpenAI：最大的生态系统，Assistants API，最佳结构化输出

Anthropic：最长上下文，最佳指令遵循，安全关键应用

Gemini：百万级上下文，多模态，Flash层级成本最低

结论

从OpenAI开始，因其成熟的SDK和生态系统。对于长文档和关键推理任务，加入Anthropic。对于需要大上下文窗口的高容量、成本敏感操作，使用Gemini Flash。