Why Meta’s Muse Spark ditched open-source AI

Meta launched Muse Spark on April 8, its first fully closed AI model, abandoning its open-source Llama strategy

Meta launched Muse Spark on April 8, its first fully closed AI model, abandoning its open-source Llama strategy. The launch marks the inaugural product from Meta Superintelligence Labs, the unit built around Alexandr Wang after Meta’s $14.3bn investment in Scale AI.

Wang said in a statement: “Nine months ago, we rebuilt our AI stack from scratch. New infrastructure, new architecture, new data pipelines. This is step one. Bigger models are already in development with plans to open-source future versions.”

Unlike Llama, Muse Spark’s weights are not publicly accessible. API access is currently by invitation only, targeting select partners. Meta has said it hopes to open-source future versions, framing the current closure as temporary.

Gartner analyst Arun Chandrasekaran described the move as a “major shift,” saying it signals Meta’s intention to move away from the Llama brand entirely.

What Muse Spark actually does

The model is natively multimodal, handling text, image, and voice inputs. Its flagship feature is a “Contemplating” mode that runs multiple reasoning agents in parallel before responding, competing directly with Gemini Deep Think and GPT Pro.

Meta collaborated with over 1,000 physicians to curate health-related training data, and the model is being marketed as a personal health reasoning tool alongside its general assistant capabilities.

Muse Spark sits below GPT-5.4 and Gemini 3.1 Pro on the Artificial Analysis Intelligence Index, scoring 52 against their 57. Meta has not disclosed the model’s parameter count or architecture details. As crypto.news reported, the model beat Gemini 3.1 Pro on several health-related benchmarks that Meta prioritized in its evaluation suite.

Why Meta made the switch now

As crypto.news documented, Meta had been signaling a phased approach to its next AI generation, keeping core components proprietary while assessing safety risks.

The switch to a fully closed first release reflects the competitive pressure from OpenAI and Anthropic, both of which have proprietary models generating billions in API revenue that Meta’s open-source approach could not capture.

Meta’s 2026 capital expenditure is guided at $115bn to $135bn, nearly double 2025 levels. Meta stock rose more than 9% on launch day, the strongest single-day response to a Meta product announcement in over two years. The developer community that built on Llama is now being asked to wait for a future open-source release with no confirmed timeline.