openrouter（OpenRouter 模型网关）

OpenRouter — One API Key, Hundreds of AI Models

OpenRouter is a unified API gateway that gives you access to 300+ AI models from 60+ providers — Anthropic, OpenAI, Google, DeepSeek, Meta, Mistral, xAI, and more — through a single, OpenAI-compatible API endpoint.

Instead of managing separate accounts, API keys, and billing for each provider, you use one API key and one credit balance. Switch models by changing a single parameter — no code changes needed.

OpenRouter powers 250,000+ apps with 4.2M+ users globally.

Base URL: https://openrouter.ai/api/v1

Why Use OpenRouter

Quick Start

1. Create an Account

Sign up at openrouter.ai and add credits on the Credits page. Credits work with any model or provider. There is no minimum purchase and credits don't expire.

2. Create an API Key

Go to Settings > API Keys and create a key. Store it securely as an environment variable:

export OPENROUTER_API_KEY="sk-or-v1-..."

3. Make a Request

OpenRouter is fully OpenAI-compatible. Here's a basic example:

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-4-20250514",
    "messages": [
      {"role": "user", "content": "Hello, what can you help me with?"}
    ]
  }'

To switch models, just change the model field — no other code changes needed.

4. Using the OpenAI SDK

from openai import OpenAI

client = OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key="sk-or-v1-..."
)

response = client.chat.completions.create(
    model="anthropic/claude-sonnet-4-20250514",
    messages=[{"role": "user", "content": "Hello!"}]
)

Pricing

OpenRouter passes through provider pricing at cost — no markup.

Input and output tokens are billed separately per model. You're only charged for successful completions — failed requests and fallback attempts don't cost you.

Sample Model Pricing (per 1M tokens)

*Prices change as providers update. Check openrouter.ai/models for current rates.*

Key Features

Model Routing & Fallbacks

OpenRouter automatically routes requests to the best available provider. If one provider goes down or errors, the request falls back to another — you're only billed for the successful run.

You can also control routing explicitly:

• Provider ordering — Prioritize specific providers for a model
• Provider ignoring — Exclude providers you don't trust
• Performance thresholds — Set latency or throughput minimums
• Regional routing — Pin requests to specific regions (Enterprise / Pay-as-you-go)
• :nitro variant — Append :nitro to a model slug for speed-optimized routing

Bring Your Own Key (BYOK)

Use your existing API keys from OpenAI, Anthropic, Google Cloud, AWS Bedrock, Azure AI, and other providers while still getting OpenRouter's unified interface, fallback routing, and analytics.

• First 1M BYOK requests per month are free
• 5% fee on subsequent usage (based on upstream provider cost)
• BYOK keys are automatically prioritized during routing
• Supports: OpenAI, Anthropic, Google Vertex, AWS Bedrock, Azure, and more

Tool Calling & Structured Outputs

If the underlying model supports function calling / tool use, you can use it through the same API. Structured output formats (JSON mode) are also supported for compatible models.

Web Search

Append :online to any model slug to enable web search (powered by Exa.ai). Search results are injected into the context with citations. Pricing is per result returned.

Multimodal

OpenRouter supports images, audio, and PDFs across models that handle them. Send images as URLs or base64, and PDFs work the same way.

Prompt Caching

Most major providers now support context caching through OpenRouter. Gemini offers a 90% discount on cached tokens.

Zero Completion Insurance

Every request includes Zero Completion Insurance — if a model returns an empty or failed completion, you're not charged.

Zero Data Retention (ZDR)

For sensitive workloads, enable ZDR to ensure providers don't retain your prompts or completions.

API Reference

All endpoints use Authorization: Bearer <OPENROUTER_API_KEY>.

Base URL: https://openrouter.ai/api/v1

Core Endpoints

Key Management Endpoints

Analytics & Account

Security & Privacy

• No training on your data. OpenRouter does not train on prompts or completions.
• Provider logging controls. Providers that log prompts are flagged. You can

restrict routing to non-logging providers in your privacy settings.

• Fine-grained data policies. Ensure prompts only go to providers you trust.
• SOC 2 Type I compliance indicated (verify current status for compliance-critical

deployments).

• ZDR (Zero Data Retention) available per-request or account-wide.
• API key spending limits — Set per-key caps with daily, weekly, or monthly resets.

Framework Integrations

OpenRouter works with popular frameworks out of the box:

• OpenAI SDK (Python & TypeScript) — change the base URL
• LangChain / LangGraph
• Vercel AI SDK
• Langfuse (observability)
• Mastra
• MCP Servers
• Any framework that supports the OpenAI API format

Organizations & Enterprise

For teams, OpenRouter offers organizations with:

• Shared credit pools and centralized billing
• Role-based access control (Admin / Member)
• Organization-wide activity tracking and CSV/PDF exports
• Guardrails — spending limits, model/provider allowlists, ZDR enforcement
• Programmatic key provisioning and rotation via Management API
• SSO (SAML) on Enterprise plans
• Broadcast / observability integrations (up to 5 destinations)

Free Models

OpenRouter offers dozens of models at zero cost, including DeepSeek R1, Llama 3.3 70B, Gemma 3, and others. Free models are rate-limited (20 req/min, 200 req/day for accounts without credits; 1,000/day with $10+ credits) but require no payment to use.

A dedicated free model router (openrouter/free) automatically selects the best available free model for your request.

Contact & Support

Important Notes

• Save your API key on creation. Store it in your platform's secrets manager or

as OPENROUTER_API_KEY in your environment.

• Never send your OpenRouter API key to any domain other than openrouter.ai.

Your key grants access to your credit balance and all enabled models.

• Model availability can change. Models can be deprecated or have pricing updates

at any time. Pin specific model IDs/versions in production to avoid surprises.

• Free model rate limits apply. Free-tier usage is subject to rate limiting,

especially during peak times. Failed attempts still count toward daily quotas.

• BYOK is not free after 1M requests. The first million BYOK requests per month

are free, then a 5% fee applies based on upstream cost.

• OpenRouter is a proxy, not a provider. It routes your requests to the underlying

model provider. Latency, output quality, and capabilities depend on the model and provider, not OpenRouter itself.