OpenRouter

What is OpenRouter?

OpenRouter is a unified API and model marketplace for language, reasoning, vision, image, speech, and agent workloads. One account and API key can access hundreds of models across many inference providers through an interface designed to work with OpenAI-compatible SDKs. Each model page exposes pricing, context, capabilities, and available providers.

The routing layer can choose providers based on price, throughput, availability, data policy, or user-defined constraints, and can fall back when a provider fails. Teams can centralize credits and usage analytics instead of maintaining a separate integration and billing relationship for every model provider.

How to Use OpenRouter

1. Create an account and add credits or select an available free model for testing.
2. Generate an API key for the application or organization.
3. Point an OpenAI-compatible SDK to OpenRouter and select a model identifier.
4. Configure routing, fallback, privacy, price, or performance constraints when needed.
5. Monitor request cost, token use, latency, and provider behavior in the activity dashboard.

Core Features

• Unified model API: Accesses models from many developers and inference providers through one interface.
• OpenAI compatibility: Works as a drop-in endpoint for many applications using OpenAI SDK request shapes.
• Provider routing and fallback: Selects eligible providers and retries through alternatives when configured.
• Price and performance controls: Routes by price, throughput, quantization, and maximum cost constraints.
• Privacy policies: Lets teams restrict providers according to logging, training, and data-retention requirements.
• Usage and billing analytics: Consolidates credits, request history, token counts, latency, and cost information.

Use Cases

• Multi-model applications: Product teams can switch or compare models without maintaining a separate adapter for each provider.
• Reliable production inference: Applications can use provider fallback to reduce the impact of individual outages.
• Cost optimization: Workloads can route to lower-cost providers or models while enforcing price ceilings.
• Agent infrastructure: Agent builders can use tool calling, structured outputs, reasoning controls, and a broad model catalog through one API.

Pricing

OpenRouter uses prepaid credits and passes through model inference prices without markup. It charges a platform fee when users purchase credits; the current published fee is 5.5%, with a minimum charge on small purchases. Free model variants are available with lower rate limits. Bring-your-own-key usage includes a monthly free request allowance, after which an additional usage fee may apply. Each model and provider has its own token or media price, so production teams should check the selected route rather than assuming one platform-wide inference rate.

Frequently Asked Questions

Is OpenRouter a model provider?

It is primarily a routing and billing layer that proxies requests to multiple model and inference providers, although it also offers interfaces and platform capabilities around those requests.

What happens when a provider is unavailable?

OpenRouter can automatically fall back to another eligible provider according to the configured routing rules.

Are prompts stored by OpenRouter?

The company states that it logs basic request metadata but does not log prompts and completions by default. Requests are still sent to selected upstream providers, so teams should configure privacy policies appropriate to their data.