The model shipped in April. The interface developers need to build on it has slipped repeatedly, with no firm date until Meta said this week it would arrive this month.
A model without an API is a demo, not a platform. That is the awkward position Meta’s Muse Spark has occupied since April, when the company launched the model but held back the application programming interface that outside developers need to build on it. According to the Wall Street Journal, Meta has pushed that release back repeatedly and, as of Tuesday, had no scheduled launch date.
The gap has now stretched close to two months. Meta’s AI chief told developers to expect the API “soon” after the April launch, the Journal reported, citing people familiar with the matter, and soon has kept receding. For developers who have built product plans around Muse Spark, an indefinite delay on the interface is the part that actually bites; the model exists, but the means of using it at scale does not.
Meta’s account is more upbeat. A company spokesperson told Reuters on Wednesday that Meta is already testing the API with some early partners and expects to release it this month. The two versions are not flatly contradictory. A limited private test running while a public launch date stays unset is exactly the situation the Journal described, read from the optimistic end.
The delay is small in isolation and larger in context. Meta has committed to AI spending on a scale that reshaped the company this spring, including thousands of job cuts justified explicitly as a shift of payroll into AI infrastructure.
Against capital expenditure measured in the tens of billions, a developer interface that ships two months late is a rounding error. It is also exactly the kind of operational detail that signals whether the spending is translating into shippable product on schedule.
Neither Meta nor the Journal’s sources gave a reason for the slips. There was no stated performance problem, no safety hold, no named technical blocker, only a date that kept moving and a model sitting in front of developers who could not yet plug into it.
The API is the difference between a model people can admire and one they can build businesses on. Without it, third-party developers cannot integrate Muse Spark into their own products, cannot call it programmatically at scale, and cannot ship anything that depends on it.
A consumer can use the model where Meta surfaces it; a developer building a product around it is left waiting. For a company that has framed its AI strategy partly around becoming a platform others build on, that is the part of the delay with the longest reach.
The slip also lands against a competitive backdrop in which rivals have shipped developer access alongside their models rather than after them. OpenAI, Anthropic, and Google have made programmatic access a launch-day feature, and the gap between a model announcement and a usable API has become a rough proxy for how production-ready an effort actually is.
Meta launching the model first and the interface months later inverts the order its competitors have set.
What comes next is narrow and testable. Meta has put itself on record that the API arrives this month. If it does, the episode is a footnote. If June ends the way April, and May did, with the model live and the interface still in private testing, the question stops being about a schedule and starts being about why one of the best-funded AI efforts in the industry cannot ship the front door to its own model.
