OpenRouter API Pricing Sheet & Gateway Rates
OpenRouter provides a single unified OpenAI-compatible endpoint that routes requests across 250+ LLMs and 40+ host clouds. Compare raw host token rates, automatic price arbitrage, and fallback economics.
OpenRouter Multi-Provider Spend Simulator
Calculate monthly bills across unified LLM gateway traffic with dynamic provider routing.
Production Pricing & Model Tier Rates
Verified rates, context capacities, typical inference latency, and recommended enterprise workloads.
| Model / Tier | Input Rate (1M) | Output Rate (1M) | Context Window | Latency / Throughput | Recommended Workload |
|---|---|---|---|---|---|
| DeepSeek-V3 (Cheapest Host) | $0.270 | $1.100 | 128,000 | Host dependent | Automated lowest-cost routing across DeepSeek, Together, Fireworks |
| Claude 3.7 Sonnet (Anthropic) | $3.000 | $15.000 | 200,000 | Anthropic direct | Flagship reasoning with dynamic fallback to Bedrock / Vertex |
| Llama 3.3 70B (Groq / Together) | $0.590–$0.880 | $0.790–$0.880 | 128,000 | 100–300 tok/s | Fast open-weight reasoning with multi-host failover |
| OpenAI GPT-4o (Direct) | $2.500 | $10.000 | 128,000 | OpenAI direct | Multimodal analysis with Azure failover |
âš¡ Architectural Deep Dive & Unit Economics Analysis
OpenRouter solves the critical enterprise operational challenges of vendor lock-in, API downtime, and price volatility. By providing a single unified gateway, engineering teams can configure multi-provider fallback chains in their request headers (e.g. provider: {{ order: ['Groq', 'Together', 'DeepInfra'] }}). If Groq encounters rate limiting or downtime, OpenRouter instantly retries on Together AI with zero application exceptions.
OpenRouter charges zero markup on host token pricing for standard API usage, passing through the exact underlying provider rates. Developers can fund their account with a single credit card, crypto, or enterprise invoice, gaining access to Anthropic, OpenAI, Google, DeepSeek, and open-source models under a unified billing ledger.
Production Workloads & Monthly Cost Modeling
Evaluating practical unit economics across real-world application architectures.
Zero-Downtime Mission-Critical SaaS Assistant
Routing prompts through OpenRouter with automatic failover across Anthropic direct, AWS Bedrock, and Google Cloud Vertex AI.
Cost-Optimized Llama 3.3 70B Dynamic Arbitrage
Configuring OpenRouter to automatically select the cheapest online provider (routing between DeepInfra, Groq, and Together AI).
Multi-Tenant Enterprise AI Gateway
Managing API keys and per-user token quotas across 50 internal developers accessing 30 different AI models with unified audit logs.
import os
from openai import OpenAI
# OpenRouter is 100% wire-compatible with the OpenAI SDK
client = OpenAI(
api_key=os.environ.get("OPENROUTER_API_KEY"),
base_url="https://openrouter.ai/api/v1"
)
def query_with_fallback(prompt: str):
"""Route to Claude 3.7 with automatic failover to DeepSeek-V3."""
response = client.chat.completions.create(
model="anthropic/claude-3.7-sonnet",
messages=[{"role": "user", "content": prompt}],
extra_headers={
"HTTP-Referer": "https://checkapicost.com",
"X-Title": "CheckAPICost Production Gateway"
},
extra_body={
"models": ["anthropic/claude-3.7-sonnet", "deepseek/deepseek-chat"],
"route": "fallback"
}
)
return response.choices[0].message.contentOperational Gotchas & Failure Modes
Engineering realities, throttling thresholds, and hidden surcharge traps to monitor.
Related Models, Showdowns & Benchmarks
Cross-reference provider rates against hosted models, direct head-to-head battles, and calculators.
OpenRouter vs Groq Showdown
Compare unified gateway flexibility against direct LPU speed.
Groq vs Together AI Showdown
Evaluate the underlying inference hosts behind OpenRouter.
Cheapest LLM APIs of 2026
Find the lowest cost token providers on the market.
Cursor vs Cline Cost Calculator
Calculate monthly IDE bills using OpenRouter API keys.
Frequently Asked Questions
Does OpenRouter charge extra markup on model token prices?
No. For standard API requests, OpenRouter passes through the exact underlying host token pricing without markup, making money primarily on enterprise routing and payment processing.
What happens if a provider goes down or rate-limits my key?
OpenRouter supports automatic fallback. In your API request, you can specify an ordered list of models or providers; if the primary host errors, OpenRouter retries the next provider within milliseconds.
Can I use the standard OpenAI Python client with OpenRouter?
Yes! OpenRouter is 100% wire-compatible with the official OpenAI SDK. Set base_url to 'https://openrouter.ai/api/v1' and supply your OPENROUTER_API_KEY.
Does OpenRouter support prompt caching?
Yes! For supported models (Anthropic Claude, DeepSeek-V3), OpenRouter supports prompt caching discounts automatically.
How many models are available on OpenRouter?
OpenRouter provides access to over 250 models, including Claude 3.7, GPT-4o, DeepSeek-V3/R1, Gemini 2.5, Llama 3.3, and Qwen 2.5.
Does OpenRouter log prompts or train on data?
OpenRouter does not train models on customer API traffic and offers a zero-retention privacy toggle that routes exclusively through providers offering zero data retention.
OpenRouter Gateway Mechanics: Provider Routing, Hedging & Token Markups
OpenRouter functions as a universal API gateway providing unified access to over 250 AI models across dozens of competing inference providers (Together, Fireworks, DeepInfra, Groq, Cerebras, and official provider APIs). The platform offers a critical architectural capability: Provider Routing & Hedging. Developers can specify provider sorting rules (e.g. `order: ["DeepInfra", "Together", "Fireworks"]`) or enable automatic latency routing (`route: "fallback"`), ensuring requests fail over seamlessly if a primary host experiences a 503 error.
OpenRouter Provider Routing Configuration
| Routing Strategy | Provider Parameter | Latency SLA | Effective Cost Multiplier |
|---|---|---|---|
| Cheapest Available Host | provider: { sort: "price" } |
Variable (180-450ms) | Lowest raw cost (No markup) |
| Lowest Latency (Fastest) | provider: { sort: "throughput" } |
Sub-120ms (Groq / Cerebras) | Premium LPU token rate |
| Strict Data Privacy (No Log) | provider: { data_collection: "deny" } |
Enterprise guaranteed | Zero data retention guarantee |
OpenRouter TypeScript Client with Provider Preferences
import OpenAI from 'openai';
const openrouter = new OpenAI({
baseURL: 'https://openrouter.ai/api/v1',
apiKey: process.env.OPENROUTER_API_KEY,
defaultHeaders: {
'HTTP-Referer': 'https://checkapicost.com',
'X-Title': 'CheckApiCost Dispatcher'
}
});
export async function routeInference(prompt: string) {
return await openrouter.chat.completions.create({
model: 'meta-llama/llama-3.3-70b-instruct',
messages: [{ role: 'user', content: prompt }],
extra_body: {
provider: { order: ['DeepInfra', 'Together'], allow_fallbacks: true }
}
});
}Engineering Q&A: OpenRouter Economics
Does OpenRouter add a markup on upstream token prices?
OpenRouter charges the exact upstream provider rates for direct inference. The platform generates revenue through payment processing fees on crypto/credit card balance deposits rather than taking arbitrary token percentage markups.
What happens if all upstream providers for a model fail?
If all primary providers fail, OpenRouter returns an HTTP 502 with structured error details specifying upstream statuses. Developers should configure fallback models (e.g. `models: ["anthropic/claude-3.5-sonnet", "deepseek/deepseek-chat"]`) to ensure multi-model failover.