Grok 4.7 把模型竞争拉回成本表：编码能力、延迟与价格该怎么验

调用成本 = 输入 token / 1,000,000 × 输入单价
         + 输出 token / 1,000,000 × 输出单价
         + 额外工具费用

有效任务成本 = 所有尝试的调用成本总和 / 成功完成的任务数

#!/usr/bin/env python3
import json
import os
import time
from pathlib import Path
from urllib.request import Request, urlopen


def load_config(name: str) -> dict:
    return {
        'name': name,
        'base_url': os.environ[f'BASE_URL_{name}'].rstrip('/'),
        'api_key': os.environ[f'API_KEY_{name}'],
        'model': os.environ[f'MODEL_{name}'],
        'input_price': float(os.environ[f'INPUT_PRICE_{name}']),
        'output_price': float(os.environ[f'OUTPUT_PRICE_{name}']),
    }


def run(config: dict, prompt: str) -> dict:
    payload = {
        'model': config['model'],
        'temperature': 0,
        'messages': [
            {
                'role': 'system',
                'content': (
                    'You are a senior software engineer. Follow every constraint, '
                    'state assumptions, and return a complete implementation.'
                ),
            },
            {'role': 'user', 'content': prompt},
        ],
    }

    request = Request(
        f"{config['base_url']}/v1/chat/completions",
        data=json.dumps(payload).encode('utf-8'),
        headers={
            'Authorization': f"Bearer {config['api_key']}",
            'Content-Type': 'application/json',
        },
        method='POST',
    )

    started = time.perf_counter()
    with urlopen(request, timeout=300) as response:
        result = json.loads(response.read().decode('utf-8'))
    elapsed = time.perf_counter() - started

    usage = result.get('usage', {})
    input_tokens = usage.get('prompt_tokens', 0)
    output_tokens = usage.get('completion_tokens', 0)
    estimated_cost = (
        input_tokens * config['input_price'] / 1_000_000
        + output_tokens * config['output_price'] / 1_000_000
    )

    return {
        'provider': config['name'],
        'model': config['model'],
        'elapsed_seconds': round(elapsed, 3),
        'input_tokens': input_tokens,
        'output_tokens': output_tokens,
        'estimated_cost': round(estimated_cost, 6),
        'answer': result['choices'][0]['message']['content'],
    }


def main() -> None:
    prompt = Path('prompt.txt').read_text(encoding='utf-8')
    reports = []

    for name in ('A', 'B'):
        report = run(load_config(name), prompt)
        reports.append(report)
        Path(f"answer-{name}.md").write_text(
            report['answer'], encoding='utf-8'
        )

    summary = [
        {key: value for key, value in report.items() if key != 'answer'}
        for report in reports
    ]
    print(json.dumps(summary, ensure_ascii=False, indent=2))


if __name__ == '__main__':
    main()

cat > prompt.txt <<'EOF'
为一个 Python 3.12 服务实现带抖动的指数退避重试器。
要求：
1. 只使用标准库；
2. 支持最大尝试次数和最大等待时间；
3. 不捕获 KeyboardInterrupt 和 SystemExit；
4. 提供完整实现及 unittest 测试；
5. 解释边界条件和时间复杂度。
EOF

export BASE_URL_A='https://provider-a.example.com'
export API_KEY_A='replace-me'
export MODEL_A='replace-with-grok-model-id'
export INPUT_PRICE_A='0.00'
export OUTPUT_PRICE_A='0.00'

export BASE_URL_B='https://provider-b.example.com'
export API_KEY_B='replace-me'
export MODEL_B='replace-with-comparison-model-id'
export INPUT_PRICE_B='0.00'
export OUTPUT_PRICE_B='0.00'

python3 compare_models.py

测试是否通过 | 人工修复分钟数 | 违反约束数 | 重试次数 | 总调用成本