Meta’s Muse Spark Wraps Up Its Open-Source AI Initiative
On April 8, Meta unveiled Muse Spark, its first entirely closed AI model, signifying a shift from its open-source Llama model strategy.
Summary
- Muse Spark, launched on April 8, is the inaugural creation from Meta Superintelligence Labs, developed by Alexandr Wang’s team after a $14.3bn partnership with Scale AI.
- This model is entirely proprietary, with no public weights available, reversing the path taken by the Llama initiative, which saw 1.2 billion downloads by early 2026.
- Upon its launch, Meta’s stock rose by 9%, with plans to integrate the model across WhatsApp, Instagram, Facebook, and Messenger in the near future.
Meta launched Muse Spark on April 8 as its first completely closed AI model, indicating a definitive move away from its previous open-source Llama approach. This launch is the first output from Meta Superintelligence Labs, a division established around Alexandr Wang following Meta’s significant $14.3bn investment in Scale AI.
Wang shared, “Nine months ago, we revamped our entire AI stack. We created new infrastructure, developed fresh architecture, and updated data pipelines. This is merely the beginning; larger models are being planned, with considerations for future open-sourcing.”
Unlike Llama, the weights for Muse Spark are not made public. Currently, access to the API is limited to select partners through an invitation-only process. Meta has signaled a potential interest in open-sourcing subsequent versions, viewing the current closed nature as a temporary situation.
According to analyst Arun Chandrasekaran from Gartner, this shift is a “significant transition,” reflecting Meta’s aim to move away from Llama branding entirely.
Functionality of Muse Spark
This model is inherently multimodal, capable of handling text, images, and voice inputs. Its notable feature, an operational mode known as “Contemplating,” utilizes multiple reasoning agents simultaneously before providing a response, positioning it in competition with Gemini Deep Think and GPT Pro.
Meta has partnered with over 1,000 healthcare professionals to curate health-specific training data, positioning the model as a personal health assistant, while also functioning as a general assistant.
On the Artificial Analysis Intelligence Index, Muse Spark scores below GPT-5.4 and Gemini 3.1 Pro, achieving a score of 52 compared to their 57. Meta has not yet disclosed specific details about the model’s parameters or architecture. However, reports from crypto.news suggest it outperformed Gemini 3.1 Pro on various health metrics outlined in Meta’s evaluation framework.
Reasons Behind Meta’s Timing
As reported by crypto.news, Meta has been signaling a gradual approach toward its next AI generation, choosing to keep key components proprietary while considering safety measures.
This shift to a fully closed initial release is perceived as a response to competitive pressures from OpenAI and Anthropic, both of which have proprietary models generating substantial API revenues—something Meta’s open-source strategy could not effectively harness.
For 2026, Meta’s capital expenditures are estimated to be between $115bn and $135bn, nearly doubling that of 2025. The stock saw a notable 9% rise on launch day, marking the largest single-day response 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 version, with no set timeline outlined.