Meta Is Building an OpenRouter Rival: An AI Task Router to Cut Compute Costs

Meta's internal lab, AAI Labs, is developing an AI task router, an OpenRouter analogue that distributes requests among more cost-efficient large models to reduce the company's compute costs. The news outlet 36Kr reported this in July 2026, citing internal documents seen by Sina Finance (新浪财经).

What is AAI Labs building

AAI Labs is building an internal dispatcher service that sends different AI tasks to the large models with the best price-to-quality ratio. According to 36Kr, the key goal is to lower overall spending on computing resources, including for code development: routine requests go to cheaper models instead of everything being pushed through the most expensive ones.

• AAI Labs is an incubator within Meta's Applied AI engineering team
• The department was officially created in March 2026
• The service is similar to OpenRouter and routes tasks between different LLMs
• The goal is to reduce compute costs, including for code development
• The source is 36Kr, citing Sina Finance (新浪财经)

How Meta's incubator works

Inside AAI Labs, Meta employees submit proposals for AI products and services for the company's internal use. The incubator belongs to the Applied AI engineering team and, according to 36Kr, was formed in March 2026 specifically as a venue for such initiatives.

Once a proposal is approved, a small dedicated team is assembled to bring the product to completion. According to the internal documents, this is also the stage at which the company assesses whether to release the service outside Meta; some mature projects may get a full public release.

"After a project is approved, the company assembles a small dedicated team to finish developing the product and evaluates whether to release it externally," reads an internal

Meta document seen by Sina Finance.

Why Meta wants its own OpenRouter

Meta is building its own model router because compute bills for working with large language models grow along with the number of internal AI tools. OpenRouter, which the project is modeled on, is a gateway that distributes requests among dozens of LLMs from different providers. A similar internal system can automatically pick, for each task, the model with the best balance of price and output quality.

For Meta, this is first and foremost a question of savings: the more code and text employees generate through AI, the more the split between cheap and expensive models affects the final infrastructure budget.

What this means

Big tech companies are starting to treat AI compute as a line item to be optimized, not just scaled up. Meta's internal OpenRouter competitor signals that routing tasks between models by price is becoming standard practice for cutting costs at scale.

Frequently asked questions

What is AAI Labs building?

AAI Labs, Meta's internal incubator, is building a router service, an OpenRouter analogue that distributes different AI tasks to more cost-efficient large models to reduce the company's compute spending.

When did Meta create AAI Labs?

According to 36Kr, the department was officially formed in March 2026 and is part of the Applied AI engineering team inside Meta.

Will the service be released outside Meta?

According to the internal documents, mature AAI Labs projects may get a public release: after development, a dedicated team assesses whether to release the product externally.

*Meta is designated an extremist organization and banned in Russia.