Los límites de tarifas determinan qué tan rápido puede buscar su agente. Este tutorial crea un punto de referencia automatizado que mide el rendimiento real, la latencia y el comportamiento de límite de velocidad en Scavio, Tavily, Exa, Brave y SerpAPI. Obtendrá una tabla comparativa que muestra las solicitudes por minuto, la latencia p50/p95 y el costo por consulta para cada proveedor.
Requisitos previos
- Python 3.8+
- solicita biblioteca
- Claves API para proveedores que desea probar
- Una clave API de Scavio de scavio.dev
Guia paso a paso
Paso 1: Establecer configuraciones de proveedor
Defina el punto final, la autenticación y los límites conocidos de cada proveedor para la evaluación comparativa.
Python
import os, requests, time, statistics
PROVIDERS = {
'scavio': {
'url': 'https://api.scavio.dev/api/v1/search',
'headers': {'x-api-key': os.environ.get('SCAVIO_API_KEY', ''), 'Content-Type': 'application/json'},
'body': lambda q: {'query': q, 'country_code': 'us'},
'method': 'POST',
'cost_per_query': 0.005,
'free_tier': '250 credits/mo',
},
'tavily': {
'url': 'https://api.tavily.com/search',
'headers': {'Content-Type': 'application/json'},
'body': lambda q: {'api_key': os.environ.get('TAVILY_API_KEY', ''), 'query': q},
'method': 'POST',
'cost_per_query': 0.01,
'free_tier': '~20 req/min limit',
},
'serpapi': {
'url': 'https://serpapi.com/search',
'headers': {},
'body': lambda q: {'q': q, 'api_key': os.environ.get('SERPAPI_KEY', '')},
'method': 'GET',
'cost_per_query': 0.025,
'free_tier': '250/mo',
},
}
TEST_QUERIES = [
'best search api 2026',
'mcp protocol agent tools',
'web scraping alternative api',
]
print(f'Benchmarking {len(PROVIDERS)} providers with {len(TEST_QUERIES)} queries each')Paso 2: Ejecute el punto de referencia de latencia
Envíe solicitudes de forma secuencial y mida el tiempo de respuesta, el estado y los encabezados de límite de velocidad.
Python
def benchmark_provider(name, config, queries):
latencies = []
errors = 0
rate_limited = 0
for q in queries:
start = time.time()
try:
if config['method'] == 'POST':
resp = requests.post(config['url'], headers=config['headers'],
json=config['body'](q), timeout=15)
else:
resp = requests.get(config['url'], headers=config['headers'],
params=config['body'](q), timeout=15)
elapsed = (time.time() - start) * 1000
if resp.status_code == 429:
rate_limited += 1
print(f' {name}: 429 rate limited on query {len(latencies)+1}')
elif resp.status_code >= 400:
errors += 1
else:
latencies.append(elapsed)
except Exception as e:
errors += 1
elapsed = (time.time() - start) * 1000
print(f' {name}: error - {str(e)[:40]}')
time.sleep(0.5) # Be respectful
return {
'name': name,
'queries': len(queries),
'successful': len(latencies),
'errors': errors,
'rate_limited': rate_limited,
'p50': statistics.median(latencies) if latencies else 0,
'p95': sorted(latencies)[int(len(latencies)*0.95)] if latencies else 0,
'avg': statistics.mean(latencies) if latencies else 0,
'cost': config['cost_per_query'],
}
results = []
for name, config in PROVIDERS.items():
if not config['headers'].get('x-api-key', config['body']('test').get('api_key', 'x')):
print(f' Skipping {name}: no API key')
continue
print(f'\nBenchmarking {name}...')
result = benchmark_provider(name, config, TEST_QUERIES)
results.append(result)Paso 3: Generar la tabla comparativa
Imprima una tabla comparativa formateada que clasifique a los proveedores por latencia y costo.
Python
def print_benchmark_table(results):
print(f'\n{"=" * 70}')
print(f'{"Provider":12} {"p50 ms":>8} {"p95 ms":>8} {"Avg ms":>8} {"OK":>4} {"429":>4} {"$/query":>8}')
print(f'{"-" * 70}')
for r in sorted(results, key=lambda x: x['p50']):
print(f'{r["name"]:12} {r["p50"]:8.0f} {r["p95"]:8.0f} {r["avg"]:8.0f} {r["successful"]:4} {r["rate_limited"]:4} {r["cost"]:8.3f}')
print(f'{"-" * 70}')
# Winner
if results:
fastest = min(results, key=lambda x: x['p50'])
cheapest = min(results, key=lambda x: x['cost'])
print(f'\n Fastest (p50): {fastest["name"]} at {fastest["p50"]:.0f}ms')
print(f' Cheapest: {cheapest["name"]} at ${cheapest["cost"]}/query')
# Monthly cost at 10K queries
print(f'\n Monthly cost at 10K queries:')
for r in sorted(results, key=lambda x: x['cost']):
monthly = r['cost'] * 10000
print(f' {r["name"]:12} ${monthly:,.0f}')
print_benchmark_table(results)Ejemplo en Python
Python
import os, requests, time
SH = {'x-api-key': os.environ['SCAVIO_API_KEY'], 'Content-Type': 'application/json'}
def bench(query):
start = time.time()
resp = requests.post('https://api.scavio.dev/api/v1/search',
headers=SH, json={'query': query, 'country_code': 'us'}, timeout=10)
ms = (time.time() - start) * 1000
print(f'{query[:30]:30} | {ms:.0f}ms | {resp.status_code}')
for q in ['search api benchmark', 'rate limit test']:
bench(q)Ejemplo en JavaScript
JavaScript
const SH = { 'x-api-key': process.env.SCAVIO_API_KEY, 'Content-Type': 'application/json' };
for (const q of ['search api benchmark', 'rate limit test']) {
const start = Date.now();
const resp = await fetch('https://api.scavio.dev/api/v1/search', {
method: 'POST', headers: SH,
body: JSON.stringify({ query: q, country_code: 'us' })
});
console.log(`${q}: ${Date.now() - start}ms | ${resp.status}`);
}Salida esperada
JSON
Benchmarking 3 providers with 3 queries each
Benchmarking scavio...
Benchmarking tavily...
Benchmarking serpapi...
======================================================================
Provider p50 ms p95 ms Avg ms OK 429 $/query
----------------------------------------------------------------------
scavio 320 450 350 3 0 0.005
tavily 580 920 640 3 0 0.010
serpapi 410 680 470 3 0 0.025
----------------------------------------------------------------------
Fastest (p50): scavio at 320ms
Cheapest: scavio at $0.005/query
Monthly cost at 10K queries:
scavio $50
tavily $100
serpapi $250