Meta’s Muse Spark Concludes Its Open-Source AI Journey
On April 8, Meta introduced Muse Spark, its inaugural fully closed AI model, marking a departure from its open-source Llama strategy.
Summary
- Muse Spark, launched on April 8, is the first creation from Meta Superintelligence Labs, engineered by Alexandr Wang’s team following a $14.3bn partnership with Scale AI.
- This model is completely proprietary with no public weights, reversing the trajectory of the Llama strategy, which achieved 1.2 billion downloads by early 2026.
- On launch day, Meta stock surged by 9%, with plans to implement the model across WhatsApp, Instagram, Facebook, and Messenger in the upcoming weeks.
Meta unveiled Muse Spark on April 8 as its first entirely closed AI model, marking a clear shift from its previously open-source Llama strategy. This launch represents the first output from Meta Superintelligence Labs, a division formed around Alexandr Wang following Meta’s substantial $14.3bn investment in Scale AI.
Wang stated, “Nine months ago, we rebuilt our entire AI stack. We developed new infrastructure, fresh architecture, and revamped data pipelines. This is just the beginning; larger models are in the pipeline, with intentions to open-source future iterations.”
In contrast to Llama, the weights for Muse Spark are not available to the public. Currently, access to the API is by invitation only, aimed at select partners. Meta has expressed a desire to potentially open-source later versions, framing the current closure as a temporary measure.
According to analyst Arun Chandrasekaran from Gartner, this transition represents a “major shift,” indicating Meta’s clear intention to distance itself from the Llama branding altogether.
Functionality of Muse Spark
This model is inherently multimodal, capable of processing text, images, and voice inputs. Its standout feature is a operational mode dubbed “Contemplating,” which runs multiple reasoning agents concurrently before delivering a response, directly competing with Gemini Deep Think and GPT Pro.
Meta has collaborated with over 1,000 medical professionals to assemble health-focused training data, promoting the model as a personal health reasoning tool while also serving as a general assistant.
On the Artificial Analysis Intelligence Index, Muse Spark ranks below GPT-5.4 and Gemini 3.1 Pro, achieving a score of 52 compared to their 57. Meta has yet to reveal details about the model’s parameters or architecture. As reported by crypto.news, it outperformed Gemini 3.1 Pro on various health metrics highlighted in Meta’s evaluation framework.
Reasons Behind Meta’s Timing
As documented by crypto.news, Meta has been hinting at a gradual approach toward its forthcoming AI generation, opting to keep core elements proprietary while evaluating safety considerations.
The transition to a fully closed initial offering is seen as a response to competitive pressures from OpenAI and Anthropic, both of which possess proprietary models generating billions in API revenue—an approach Meta’s open-source strategy was unable to effectively capitalize on.
For 2026, Meta’s capital expenditures are projected between $115bn to $135bn, nearly double that of 2025. The stock experienced a notable 9% uptick on the day of the launch, representing the most significant single-day reaction to a product announcement from Meta in over two years. The developer community that once thrived on Llama must now await a future open-source release, with no timeline currently established.