Meta's Muse Spark API Is Still Missing — Two Delays In and No Launch Date
Meta launched its first closed-source AI model, Muse Spark, in April 2026 but has repeatedly delayed the developer API, leaving zero external access to the model more than six weeks after its debut. The Wall Street Journal reported a second delay in early June, with Meta saying the API is in partner testing and a June release is still intended.
Meta announced Muse Spark in April 2026 as its first fully closed-source AI model — no downloadable weights, no open release. The bet was that a proprietary frontier model with a developer API would let Meta compete directly with OpenAI, Anthropic, and Google in the enterprise and developer market. More than six weeks later, there is still no public API.
The Wall Street Journal reported in early June that Meta has missed two internal targets for API availability. The first delay pushed the launch from April into May citing bugs and infrastructure instability. The second pushed it from May into June. As of the WSJ’s report, no firm date existed. A Meta spokesperson told Reuters the API is in testing with select early partners and a June launch is still the target, but the company declined to give a date.
The timeline matters. Muse Spark is a closed model. Unlike Meta’s Llama series — which anyone can download and run — Muse Spark only exists as a hosted API. With no API, the model is functionally inaccessible to every developer outside Meta’s early partner program. OpenAI, Anthropic, and Google all ship API access on the day a model is announced. The six-week gap is not a minor delay — it is six weeks of developer mindshare going to alternatives.
Meta’s competitive position is under pressure in ways the API delay makes worse. The company raised its 2026 capital expenditure guidance to $145 billion — up from a prior range of $115–135 billion — to fund the AI infrastructure buildout. It simultaneously launched paid AI subscription tiers ($7.99 and $19.99 per month) in May 2026, trying to establish a consumer revenue stream for AI products. The stock fell more than 5% in after-hours trading after the capex revision.
The acquisition of Alexandr Wang’s Scale AI team was meant to accelerate model development. Meta spent billions bringing in one of the top AI training infrastructure experts in the world. The Muse Spark delays suggest the bottleneck is not talent or capital — it is the operational complexity of deploying a model at Meta’s scale with enterprise-grade reliability, without the years of API infrastructure experience that OpenAI and Anthropic have accumulated.
For developers evaluating AI APIs right now, the practical takeaway is simple: Muse Spark is not a real option until an API actually ships. If the launch happens in June as promised, the delay will look like a standard infrastructure growing pain. If July arrives without it, the story becomes about whether Meta can reliably execute on its AI developer platform ambitions at all.