OpenRouter

OpenRouter is a pay-as-you-go AI infrastructure gateway and model marketplace with selected free models, prepaid USD credits, and per-model token pricing, built for developers who want one OpenAI-compatible API, one account, and one billing surface for accessing many large language models across multiple providers. The service is useful for product teams, agent builders, AI application developers, and organizations that need model choice, fallback routing, cost controls, and governance without maintaining separate direct integrations for every model vendor. Five concrete facts define the product: it exposes an OpenAI-compatible endpoint that can work with OpenAI-style SDK integrations; its catalog is positioned around access to 300+ models from 50+ providers; it supports major model families such as GPT, Claude, Gemini, DeepSeek, Llama, Mistral, and xAI models; paid usage is charged from an OpenRouter credit balance according to the selected model route, input tokens, output tokens, and any cache or modality-specific pricing; and its routing layer can apply provider preferences, fallbacks, price ceilings, and load balancing so an application can shift traffic when a route is unavailable or too expensive. Pricing is not a single flat subscription because every model has its own live rate card. As examples of the kind of buyer-visible rates teams must compare, Gemini 2.5 Flash is listed at $0.30 per 1M input tokens and $2.50 per 1M output tokens, Claude Sonnet 4.5 is listed at $3 per 1M input tokens and $15 per 1M output tokens for standard context, and OpenAI GPT-5 provider listings show $1.25 per 1M input tokens and $10 per 1M output tokens. OpenRouter also states that pricing shown in the model catalog is what customers pay, with provider pricing passed through rather than hidden behind a universal markup. This makes the platform attractive when teams need transparent model comparison, but it also means production forecasting requires workload-specific math: prompt length, completion length, context reuse, provider route, fallback behavior, cache use, and model mix all affect the bill. For free experimentation, selected :free models are available with rate limits; for production, teams top up credits and spend them across supported models; for larger organizations, OpenRouter presents enterprise buying around volume, prepayment credits, annual commits, workspace controls, governance, data policies, and procurement needs. The operational value is strongest when an application benefits from model diversity. A SaaS assistant can use a cheaper model for routine classification, a premium reasoning model for difficult tasks, and fallback providers for customer-facing reliability while keeping the application code close to one OpenAI-compatible integration. Governance features such as custom data policies and provider restrictions matter for teams that need to control where prompts are sent. The main tradeoff is that OpenRouter adds a gateway dependency between the application and the upstream model provider, and some provider-specific capabilities, beta flags, commercial terms, or latency optimizations may still be better handled through direct vendor integrations.