OpenRouter Inc., a startup working to ease the development of artificial intelligence applications, today announced that it has secured $40 million in funding.

The company raised the capital over two rounds. They were led by Andreessen Horowitz and Menlo Ventures with participation from Sequoia Capital, as well as multiple angel investors.

AI providers such as OpenAI make their algorithms available to developers through application programming interfaces, or APIs. API designs vary significantly across providers. As a result, developers have to learn the API of each AI model they plan to use, which can require a significant time investment. 

OpenRouter promises to ease the task. It provides a cloud platform that enables developers to access AI models from more than 60 different companies through a single API. As a result, software teams don’t have to familiarize themselves with each AI provider’s API architecture.

The company says its platform also offers other benefits. Switching an application from one AI provider to another requires adapting it to the new provider’s API. That can require extensive code changes, which slows down software development and increases costs. OpenRouter’s platform removes the need for such modifications because developers can keep using its API after switching to a new AI model.

Billing is another task that the company promises to ease. If an AI application uses large language models from three different providers, the application’s developer usually has to sign three different procurement contracts. OpenRouter provides a consolidated billing tool that allows developers to buy inference credits and use them across multiple providers.

The company’s platform includes failover features designed to avoid application downtime. If the AI model on which a workload relies to work goes offline, OpenRouter automatically reroutes requests to another algorithm.

Users can customize how the platform picks the AI model that should process a given prompt. For example, a developer could specify that complicated requests should only be sent to an LLM with reasoning features. Prompts that contain a significant amount of data, in turn, can be routed to an AI with a large context window.

OpenRouter’s customization features also lend themselves to lowering inference expenses. According to the company, developers can configure its platform to prioritize the most cost-efficient LLMs when determining how to process queries. A caching feature makes it possible to reuses common prompt responses. That removes the need to generate them from scratch each time, which avoids the associated inference costs.

Alongside the standard version of its API, OpenRouter offers an enterprise edition with additional features. It provides expanded technical support, high rate limits and optimizations designed to reduce AI applications’ latency.

The company claims it processes 8.4 trillion tokens per month for customers. The annualized inference spending managed by the company’s platform topped $100 million this past April, which amounts to a tenfold increase in six months.

OpenRouter will invest the new capital in product development initiatives. It plans to expand the number of AI models supported by its platform, which currently provides access to more than 400 algorithms. The company also plans to enhance the platform’s enterprise edition. 

Image: Unsplash