技能 编程开发 AI模型可用性与健康检查

AI模型可用性与健康检查

v20260423
openrouter-model-availability
本技能用于监控通过OpenRouter接入的各类大模型(如GPT-4o, Claude)的实时状态和健康状况。它提供主动健康探测、模型目录查询和定时监控脚本,确保系统在关键模型出现故障、性能下降或不可用时,能及时发现问题并触发故障转移机制,适用于构建高可靠性的AI应用。
获取技能
292 次下载
概览

OpenRouter Model Availability

Overview

OpenRouter's /api/v1/models endpoint is the source of truth for model availability. Models can be temporarily unavailable, have degraded performance, or be permanently removed. This skill covers querying model status, building health probes, tracking availability over time, and automating failover.

Query Model Status

# Check if specific models exist and their status
curl -s https://openrouter.ai/api/v1/models | jq '[.data[] | select(
  .id == "anthropic/claude-3.5-sonnet" or
  .id == "openai/gpt-4o" or
  .id == "openai/gpt-4o-mini"
) | {
  id,
  context_length,
  prompt_per_M: ((.pricing.prompt | tonumber) * 1000000),
  completion_per_M: ((.pricing.completion | tonumber) * 1000000)
}]'

# List all available models (just IDs)
curl -s https://openrouter.ai/api/v1/models | jq '[.data[].id] | sort'

# Count models by provider
curl -s https://openrouter.ai/api/v1/models | jq '[.data[].id | split("/")[0]] | group_by(.) | map({provider: .[0], count: length}) | sort_by(-.count)'

Health Check Service

import os, time, logging
from datetime import datetime, timezone
from dataclasses import dataclass
import requests
from openai import OpenAI, APIError, APITimeoutError

log = logging.getLogger("openrouter.health")

@dataclass
class HealthStatus:
    model: str
    available: bool
    latency_ms: float
    checked_at: str
    error: str = ""

client = OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key=os.environ["OPENROUTER_API_KEY"],
    timeout=15.0,
    default_headers={"HTTP-Referer": "https://my-app.com", "X-Title": "health-check"},
)

def probe_model(model_id: str) -> HealthStatus:
    """Send a minimal request to test model availability."""
    start = time.monotonic()
    try:
        response = client.chat.completions.create(
            model=model_id,
            messages=[{"role": "user", "content": "hi"}],
            max_tokens=1,  # Minimal cost
        )
        latency = (time.monotonic() - start) * 1000
        return HealthStatus(
            model=model_id, available=True, latency_ms=round(latency, 1),
            checked_at=datetime.now(timezone.utc).isoformat(),
        )
    except (APIError, APITimeoutError) as e:
        latency = (time.monotonic() - start) * 1000
        return HealthStatus(
            model=model_id, available=False, latency_ms=round(latency, 1),
            checked_at=datetime.now(timezone.utc).isoformat(),
            error=str(e),
        )

def check_critical_models() -> list[HealthStatus]:
    """Probe all critical models."""
    CRITICAL_MODELS = [
        "anthropic/claude-3.5-sonnet",
        "openai/gpt-4o",
        "openai/gpt-4o-mini",
        "google/gemini-2.0-flash-001",
    ]
    results = []
    for model in CRITICAL_MODELS:
        status = probe_model(model)
        log.info(f"{'OK' if status.available else 'FAIL'} {model} ({status.latency_ms}ms)")
        results.append(status)
    return results

Catalog-Based Availability Check

def check_model_exists(model_id: str) -> dict:
    """Check if a model exists in the catalog (no API call cost)."""
    resp = requests.get("https://openrouter.ai/api/v1/models")
    models = {m["id"]: m for m in resp.json()["data"]}

    if model_id in models:
        m = models[model_id]
        return {
            "exists": True,
            "context_length": m["context_length"],
            "pricing": m["pricing"],
        }
    return {"exists": False, "suggestion": find_similar(model_id, models)}

def find_similar(model_id: str, models: dict) -> list[str]:
    """Find models with similar names (for migration when model is removed)."""
    prefix = model_id.split("/")[0]
    return [m for m in models if m.startswith(prefix)][:5]

Availability Monitoring Script

#!/bin/bash
# Run as cron job: */5 * * * * /path/to/check_models.sh

MODELS=("anthropic/claude-3.5-sonnet" "openai/gpt-4o" "openai/gpt-4o-mini")
LOG_FILE="/var/log/openrouter-health.log"

for MODEL in "${MODELS[@]}"; do
  START=$(date +%s%N)
  HTTP_CODE=$(curl -s -o /dev/null -w "%{http_code}" \
    https://openrouter.ai/api/v1/chat/completions \
    -H "Authorization: Bearer $OPENROUTER_API_KEY" \
    -H "Content-Type: application/json" \
    -d "{\"model\":\"$MODEL\",\"messages\":[{\"role\":\"user\",\"content\":\"ping\"}],\"max_tokens\":1}" \
    --max-time 15)
  END=$(date +%s%N)
  LATENCY=$(( (END - START) / 1000000 ))

  STATUS="OK"
  [ "$HTTP_CODE" != "200" ] && STATUS="FAIL"

  echo "$(date -u +%Y-%m-%dT%H:%M:%SZ) $STATUS $MODEL $HTTP_CODE ${LATENCY}ms" >> "$LOG_FILE"
done

Error Handling

Error Cause Fix
Model not in catalog Model renamed or removed Use find_similar() to find replacement
Health check timeout (>15s) Model overloaded or cold-starting Distinguish slow vs down; increase timeout for probes
False positive down Transient network issue Require 2-3 consecutive failures before alerting
402 on health check Credits exhausted Health checks cost ~$0.0001 each; ensure adequate credits

Enterprise Considerations

  • Health probes cost tokens ($0.0001 or less per probe with max_tokens: 1) -- budget for monitoring
  • Require 2-3 consecutive failures before marking a model as down to avoid false positives
  • Cache the models list and refresh every 5 minutes -- don't hit /api/v1/models on every request
  • Subscribe to OpenRouter announcements for model deprecations and new additions
  • Maintain a model alias map so your code uses logical names (e.g., "primary-chat") that you can remap
  • Alert when critical models disappear from the catalog, not just when they fail probes

References

信息
Category 编程开发
Name openrouter-model-availability
版本 v20260423
大小 9.21KB
更新时间 2026-04-28
语言