APIRank
EN 中文

OpenRouter

Listed at https://openrouter.ai

Overall Rank #3 ⭐⭐⭐ Highly Recommended
⚠️ Partial (not blocked in China but unstable) | 🌍 International

💰 Token Pricing

TypePriceNote
Input $0.065/M to $15/M 取决于模型，平均 $0.50/M per million tokens
Output $0.26/M to $75/M 取决于模型，平均 $2/M per million tokens
💡 Free Credits:

🤖 Supported Models (300)

GPT-4oClaude Opus/Sonnet/HaikuDeepSeek V3/R1Gemini 2.0/2.5 ProLlama 3.2/3.3 70B/405BMistral Large/SmallQwen 2.5 72B/32BGrok-2Cohere Command R+Perplexity SonarGrok-3 Beta

✨ Pros

  • ✓Gateway to 400+ models from 70+ providers through a single OpenAI-compatible API endpoint
  • ✓Intelligent fallback routing: auto-retry, provider failover, and latency-optimized provider selection
  • ✓Live leaderboards, rankings, and market share data across text, image, embeddings, and audio modalities
  • ✓Pay-as-you-go credits with no monthly subscription — top up and use any model instantly
  • ✓Advanced routing features: model fallbacks, provider selection rules, auto-exacto mode, and sovereign AI
  • ✓Zero Completion Insurance and Zero Data Retention options for enterprises needing data safety guarantees
  • ✓Rich ecosystem: 250k+ apps and 4.2M+ global users including Replit, Hermes Agent, and Kilo Code

⚠️ Cons

  • ×Model routing obfuscates original provider for debugging
  • ×China access requires proxy, connection is unstable
  • ×Aggregated API latency 15-30% higher than direct provider endpoints
  • ×Provider selection is non-deterministic, less control for advanced users
  • ×Advanced features (fallback/routing/zero-dataretention) require paid tiers

🎯 Best For

Developers who need instant access to hundreds of models via one API; teams building multi-provider redundancy into production apps; AI researchers comparing model performance; indie developers and startups wanting no-commitment pay-as-you-go access to frontier models

📊 Pricing & Value

OpenRouter uses a credit-based pay-as-you-go model — no monthly subscriptions, no commitments. You buy credits ($10, $50, $99+ options), and spend them on any of 400+ models. Pricing is transparent: the Models page shows exact per-M-token costs for both input and output across all models. OpenRouter adds a small markup (typically 0-30% via 'App Routing') on top of the raw provider rate to cover routing infrastructure and failover guarantees. The platform publishes live pricing data that updates in real-time. For budget-conscious users, the leaderboards allow filtering by value (price-to-quality ratio), making it easy to find cost-effective alternatives to expensive frontier models. Free models (like NVIDIA Nemotron, Ling-3.0-flash) are available and consume zero credits.

🔧 API & Developer Experience

  • •SDK Compatibility: Fully OpenAI-compatible — use any OpenAI SDK (Python, Node, Go, etc.) by changing the base URL to openrouter.ai/api/v1
  • •Base URL: https://openrouter.ai/api/v1 — drop-in replacement for https://api.openai.com/v1
  • •Authentication: Standard Bearer token via API key (OPENROUTER_API_KEY). Create keys via the dashboard with per-key credit limits
  • •Retry & Fallback: Built-in automatic retry on rate limits/errors; model fallback chains let you specify backup models
  • •Structured Outputs: Full support via JSON mode and structured output APIs — compatible with OpenAI's response_format
  • •Provider Selection: Manual provider pinning, auto-exacto mode, or automatic routing by price/latency/availability

🌐 Regional Availability & Latency

OpenRouter's API endpoint is not blocked in mainland China, but connection reliability is inconsistent — users typically need a proxy or VPN for stable access. Latency from China is noticeably higher than direct API calls to domestic providers (DeepSeek, Kimi, etc.) or China-optimized aggregators like FreeModel. For production workloads serving China-based users, a domestic provider or self-hosted solution is recommended. For developers outside China, OpenRouter offers excellent global performance with points-of-presence optimized for low-latency routing. The platform runs at the edge, routing requests through the fastest available upstream provider.

🔐 Enterprise & Privacy Features

OpenRouter offers enterprise-grade features including: Zero Data Retention (data never stored, no training usage), Zero Completion Insurance (no output caching), Sovereign AI routing (keep data within specific jurisdictions), Single Sign-On (SSO) with SCIM group mappings, workspace budgets with granular spend controls, and custom data policies that restrict which providers can process prompts. These features make OpenRouter viable for regulated industries (healthcare, finance, legal) where data sovereignty and compliance are critical.