Meta just stopped playing catch-up and decided to build something different. On April 8, 2026, the company unveiled Muse Spark — the first model from its newly formed Meta Superintelligence Labs — and it’s proprietary, not open source, which is a sharp departure from everything Meta has done before.
This isn’t a minor model update. Muse Spark is Meta’s first reasoning model — meaning it works through problems step by step, trying alternate strategies if the first approach fails. Every previous Meta model was built to produce an instant answer based on training data. This is a fundamentally different architecture.
So what changed? Everything, essentially.
What Exactly Is Meta Muse Spark?
Muse Spark is a natively multimodal reasoning model with support for tool use, visual chain of thought, and multi-agent orchestration. It’s the first product of a complete ground-up rebuild of Meta’s AI efforts.
Think of it this way — most AI assistants read what you type. Muse Spark can look at what you’re looking at. Snap a photo of an airport snack shelf and Meta AI powered by Muse Spark can identify and rank snacks by protein content. Scan a product and ask how it stacks up against alternatives. It’s the difference between an AI that waits for you to explain the world and one that simply looks at the world with you.
That framing matters. Meta isn’t positioning this as a research model. It’s positioning it as a personal intelligence layer embedded directly into daily life.
The Three Modes Nobody’s Talking About
Most coverage focuses on benchmarks. What’s actually more interesting for everyday users is how Muse Spark handles different types of questions.
Muse Spark offers three distinct modes: Instant mode for fast responses, Thinking mode for step-by-step reasoning, and Contemplating mode for complex tasks that require multi-agent parallel reasoning.
That last one is significant. Users can now alternate between modes depending on the sophistication of their prompt — tap one mode for a quick answer, switch to another when analyzing a legal document or extracting nutritional information from a grocery photo.
Most users won’t even realize they’re using a different architecture. That’s by design.
Who’s Behind This — And Why That Matters
In June 2025, Meta spent $14.3 billion to acquire a 49% nonvoting stake in Scale AI and brought in its co-founder and CEO, Alexandr Wang, as Meta’s first-ever chief AI officer. Wang now leads Meta Superintelligence Labs and oversees the entire Muse rollout.
Nine months. That’s how long Wang’s team had to rebuild Meta’s AI stack from scratch.
Over that period, Meta Superintelligence Labs rebuilt its AI stack from the ground up, moving faster than any development cycle it had run before. The result was Muse Spark — originally code-named Avocado internally.
The context around Llama 4 is worth understanding here. Meta’s decision to delay Muse Spark from its March timeline until May — and then release in April — allowed the company to narrow the competitive gap with Gemini 3.1 Pro and GPT-5.4 rather than shipping an underperforming model. That patience, unusual for the AI industry, paid off in market terms. Meta stock surged nearly 10% over five trading days following the announcement.
Where Muse Spark Actually Wins (And Where It Doesn’t)
Be honest about benchmarks: Muse Spark does not lead across the board.
It scores 52 on the Intelligence Index, trailing leaders Gemini 3.1 Pro and GPT-5.4 (both at 57) and Claude Opus 4.6 (53). In coding tasks specifically, Meta openly acknowledges a gap.
But here’s where the story gets genuinely interesting — health reasoning.
On HealthBench Hard, Muse Spark scores 42.8 — nearly triple Gemini’s 20.6 and Claude Opus 4.6’s 14.8. That’s not a marginal lead. Meta collaborated with over 1,000 physicians to curate training data specifically for health reasoning, enabling more factual and comprehensive medical responses.
For a company with 3.2 billion daily users across its apps, health is a massive use case. Most of those users will never run a coding benchmark. But they do take photos of food labels, ask about medications, and wonder about symptoms.
Where It’s Actually Rolling Out
Muse Spark currently powers the Meta AI assistant in the standalone Meta AI app and desktop website. In the coming weeks, it will expand to Facebook, Instagram, WhatsApp, Messenger, and the Ray-Ban Meta AI glasses. Meta also plans for Muse Spark to eventually power the Vibes AI video feature.
That distribution footprint is something no other AI lab can replicate. OpenAI doesn’t have 3 billion users. Anthropic doesn’t have WhatsApp. Google has reach, but not the same social context layer.
All versions of the model are free to use, though Meta may impose rate limits. Consumers should note that Meta’s privacy policy sets few limits on how the company can use data shared with its AI systems.
That last part deserves attention, especially as the model handles health-related queries.
The Open Source Question
This is where Meta’s strategy gets murky. The Llama family was celebrated by the developer community precisely because it was open. Muse Spark is not.
The model is proprietary, with Meta expressing only “hope” to open-source future versions. The company is also experimenting with offering third-party developers access via an API.
Meta’s AI-related capital expenditure for 2026 is projected between $115 billion and $135 billion — nearly double last year’s figure. When you’re spending at that scale, open-sourcing your flagship model immediately doesn’t make financial sense. The tension between Meta’s open-source brand identity and its new commercial ambitions is real, and unresolved.
What Comes Next
Muse Spark is explicitly described as the first step on Meta’s scaling ladder. The next generation is already in development.
The real signal here is that Muse Spark looks less like a standalone demo model and more like the operating intelligence layer Meta wants underneath its entire AI product strategy.
The AI race in 2026 is not just about who builds the smartest model. It’s about who can embed intelligence into the most touchpoints of daily life. Meta has 3.2 billion daily users, a health-optimized reasoning model, a $130 billion infrastructure commitment, and a new leadership team that rebuilt everything in nine months.
Whether Muse Spark closes the gap with OpenAI and Google on raw capability is the wrong question. The better question is whether Meta needed to close that gap — or just needed to be good enough, everywhere, for everyone.
Google Gemma 4 Is Here — And It Runs On Your Laptop, Not Just Google’s Servers
