Meta Abandons Open Source — Muse Spark Is Its First Secret AI Model

For three years, Meta's AI strategy rested on a single, distinctive pillar: open source. While Google kept Gemini closed, while OpenAI kept GPT proprietary, while Anthropic kept Claude behind a paywall, Meta released Llama — its frontier model — for anyone to download, modify and deploy. It was a competitive weapon disguised as a philosophy.
That philosophy may now be ending.

Muse Spark — Meta's First Closed Model

Meta unveiled Muse Spark — its first flagship large language model built under Chief AI Officer Alexandr Wang's newly formed Superintelligence Labs. The proprietary model — a departure from Meta's open-source Llama strategy — delivers competitive performance on multimodal perception, reasoning, health and agentic tasks at a fraction of the compute cost of its older Llama 4 mid-size variant.

Muse Spark is not open source. It is not available for download. It is not a Llama variant. It is Meta's first genuinely proprietary frontier model — built in secret, released as a closed product, and positioned as the foundation for Meta's commercial AI ambitions.

Why This Is a Strategic Reversal

Meta's open-source strategy was built on a specific competitive logic: if the most capable AI model is freely available, no competitor can use model superiority as a moat. By releasing Llama, Meta forced Google, OpenAI and Anthropic to compete on products and distribution rather than on raw model capability.
Muse Spark reverses that logic entirely. By building a closed, proprietary model, Meta is signalling that it now believes it can compete on model capability — and that it has something worth keeping secret.
Meta simultaneously announced AI capital expenditures of $145 billion for 2026 — nearly double last year's spending — signalling an aggressive push to close the gap with OpenAI and Google.

The combination is telling: $145 billion in infrastructure, a secret frontier model, and a new Superintelligence Labs unit reporting to a new Chief AI Officer. Meta is not repositioning. It is rebuilding its AI strategy from the ground up.

What Happened to Llama?

Llama is not dead. Meta has not announced the end of its open-source programme. But the launch of Muse Spark creates a two-tier Meta AI strategy that did not exist before: Llama for the open-source community, Muse Spark for commercial applications where Meta wants to capture value rather than distribute it.
The developer community that built on Llama — and that gave Meta enormous goodwill and ecosystem momentum — is watching this transition carefully. If Muse Spark becomes Meta's primary AI product and Llama becomes a secondary, maintained-but-not-frontier offering, the open-source advantage that distinguished Meta from every other GAFAM company disappears overnight.

What This Means for GAFAM

The GAFAM AI model landscape is now fully closed. Google keeps Gemini proprietary — with Gemma as its open-source offering. Microsoft keeps its OpenAI-powered models closed. Amazon keeps its Titan and Nova models proprietary. Apple keeps its foundation models entirely closed. And now Meta, with Muse Spark, has joined the proprietary camp for its flagship product.
The open-source AI movement — which Llama powered and which gave European companies and researchers an alternative to American proprietary models — has lost its most powerful corporate champion.

The European Perspective