Meta Platforms has repeatedly delayed plans to release its latest artificial intelligence (AI) model to developers, and as of Tuesday, the company had still not set a release date, according to people familiar with the matter. Nearly two months have passed since the company's head of AI told developers that the model would be released "soon." The matter has been repeatedly delayed, raising questions about whether Meta can quickly achieve commercial realization after spending huge sums of money to build its own cutting-edge AI model.
The company has been developing application programming interfaces, or APIs, software tools that allow different programs to communicate with each other. With Meta's API, applications written for computers or mobile phones will be able to run based on Meta's AI technology.
A Meta spokesperson said Wednesday that the company is testing the API with partners and plans to release it this month. "We know people need this API and we're excited to get it into their hands," the spokesperson said.
For so-called "closed source" AI models that are kept as proprietary technology and are not available for public download, the only way for developers to access them is through APIs. Enterprises usually launch corresponding APIs at the same time as the release of new models or within a few weeks of release to maximize their attention in the developer circle.
OpenAI and Anthropic make revenue in part by selling access to their APIs, which customers use to embed their AI into custom projects and tools, eliminating the hassle of building their own models from scratch.
Meta initially planned to release the API in April when it launched its latest AI model, Muse Spark, some people familiar with the matter said. Two days after the model was released, Meta Chief AI Officer Alexandr Wang posted on X that the API would be launched soon.
“The Muse Spark API is coming soon! We’re excited by the enthusiasm from developers eager to try Muse Spark in their own agent suites,” he wrote. “Stay tuned!”
However, this API has not yet been released. According to people familiar with the matter, the launch was initially delayed from April to May because of vulnerabilities discovered during testing and the need to build more infrastructure. The release was then pushed back to June, one of the people said.
Meta plans to invest up to $145 billion in capital expenditures this year, mainly to promote its AI infrastructure. The company has an ambitious vision to build personal and business intelligence for 3.5 billion daily active users, and is building AI models to achieve these goals.
In April this year, Wall Street did not buy into Meta's cash-burning plan this year that exceeded previous expectations, causing its stock price to plummet more than 5% in after-hours trading.
Meta has begun to disclose its strategy to recoup some of its costs. Last week, the company announced new subscription services for Instagram, WhatsApp and Facebook and said it would begin testing subscription services for its AI chatbot Meta AI.
Meta CEO Mark. Mark Zuckerberg said in a separate call with shareholders that creating a cloud computing business is "definitely on the table" as a means of commercially monetizing any excess computing power it builds.
Zuckerberg also said that companies approach Meta every week and ask to build API services. But he did not provide an update on when Meta might launch the Muse Spark API.
It was previously reported that Meta delayed the release of one of its AI models last year because engineers struggled to significantly improve its performance. The model, called Behemoth, was never released.
Turning around, Meta launched a large-scale recruitment and comprehensively reorganized the AI team, appointing Wang Tao as the head of the newly established Meta Superintelligence Labs. A secret division of the lab called TBD Lab developed the model that became Muse Spark.
Before Wang Tao's arrival, all AI models released by Meta were open source models, which meant developers could download and use them for free. Muse Spark, which powers Meta's AI chatbot and the AI capabilities within it, is the company's first model to launch without sharing its underlying blueprints and software files.
According to Meta's own internal benchmark evaluation, the model has the strength to compete with OpenAI and Anthropic, and scores well above xAI's Grok in most tests. However, apart from a few third-party evaluation companies that are allowed to run tests on Meta and provide ratings before the model is released, developers generally have no access to the model on their own.