145 billion expenditure pressure: Meta repeatedly delays Muse Spark API, urgent need for closed source monetization According to Beating monitoring, Meta (META) has repeatedly postponed the API opening plan for its latest AI model Muse Spark, and the release date has not been determined yet. The spokesperson stated on Wednesday that they are testing the API with partners and plan to launch it in June 2026. The delayed release has directly raised doubts from the outside world about Meta's ability to monetize huge AI investments. Meta's capital expenditure plan for this year is as high as $145 billion, and in April 2026, the stock price plummeted by over 5% due to unexpected expenses. As Meta's first closed source model, Muse Spark refuses to disclose design blueprints and software weights, and developers can only make API calls. In internal benchmark tests, Muse Spark's performance is on par with OpenAI and Anthropic's flagship models, and has surpassed xAI's Grok in most tests. However, except for a few third-party evaluation agencies, developers are still unable to test in the Agenetic Harness framework. Due to the inability to sell API interfaces, Meta can only launch paid subscriptions on Instagram, WhatsApp, Facebook, and Meta AI bots to share costs. CEO Mark Zuckerberg revealed that he does not rule out renting out excess computing power in the form of cloud services in the future. Before Muse Spark, Meta cancelled publishing the Behemoth model due to performance problems, then adjusted the AI department and established the Super Intelligent Laboratory (MSL) led by Alexandr Wang. Muse Spark is the result of the research and development of TBD Lab, the confidential department in the laboratory. [Original link]