On April 8, 2026, Meta launched Muse Spark — its most powerful AI model yet and the first from its new Meta Superintelligence Labs. After a disappointing run with Llama 4, Meta is back with a completely rebuilt AI stack. Here's what Muse Spark actually does, how it compares to the competition, and what it means for the 3 billion people who use WhatsApp, Instagram, and Facebook every day.
Muse Spark is the first model in Meta's new "Muse" series, developed by Meta Superintelligence Labs — the team assembled under Chief AI Officer Alexandr Wang, who joined Meta from Scale AI nine months ago in a deal valued at around $14 billion.
This is a significant departure from Meta's previous approach. While the company built its reputation on open-source Llama models that anyone could download and modify, Muse Spark is a closed model — its architecture and weights won't be made public, at least for now. Meta says it hopes to open-source future versions.
💡 Why the name change? The project was internally code-named "Avocado." Meta rebranded the broader series "Muse" and the first model "Muse Spark" — signaling a new chapter and distancing itself from the Llama brand after Llama 4 underperformed.
Switch between two modes depending on your task. Instant mode delivers fast answers for everyday questions. Thinking mode engages deeper reasoning for complex problems — similar to how GPT-5.4 handles extended thinking tasks.
Muse Spark can launch multiple sub-agents in parallel to tackle complex queries simultaneously. Meta's example: ask it to plan a family trip to Florida, and one agent drafts the itinerary, another compares destinations, and a third finds kid-friendly activities — all at the same time.
Snap a photo and Meta AI can analyze it in real-time context. Point it at an airport snack shelf and it identifies and ranks the highest-protein options. This becomes especially powerful on Meta's AI glasses, where the model can process what you're looking at continuously.
A new commerce feature that draws from creator content, brand posts, and community inspiration across Instagram and Facebook to help users buy clothes, decorate rooms, or discover products — with recommendations tied directly to content people already follow.
Starting now on the Meta AI app and meta.ai. Rolling out in coming weeks to WhatsApp, Instagram, Facebook, Messenger, and Meta's AI glasses — replacing the existing Llama-powered chatbots across all platforms.
The most useful independent ranking comes from the Artificial Analysis Intelligence Index v4.0, which scored Muse Spark at launch:
| Model | Intelligence Index | Strengths | Weaknesses |
|---|---|---|---|
| GPT-5.4 | 57 LEADER | Breadth, coding, reasoning | Cost at scale |
| Gemini 3.1 Pro | 57 LEADER | Reasoning, value, Google integration | Context coherence |
| Claude Opus 4.6 | 53 | Writing, long-document, accuracy | Speed on simple tasks |
| Muse Spark | 52 | Multimodal, health reasoning, social context | Abstract reasoning, coding |
The honest picture: Muse Spark is competitive but not yet at the top. It excels at figure understanding (86.4% on CharXiv Reasoning) and medical reasoning (42.8% on HealthBench Hard). But it lags in abstract reasoning (42.5% on ARC-AGI-2) and struggles with coding compared to GPT-5.4 and Claude.
That said, benchmark scores aren't the full story for Meta AI. Muse Spark's real competitive advantage isn't raw intelligence — it's distribution. No other AI model is embedded in apps used by 3+ billion people daily. When Muse Spark rolls out to WhatsApp, it will instantly become one of the most widely-used AI models in the world, regardless of how it scores on benchmarks.
Meta's Llama 4, released in early 2025, was widely considered a disappointing step. It failed to match GPT-4o or Claude 3.5 on key benchmarks and didn't capture developer enthusiasm the way Llama 3 had. That failure led Zuckerberg to bring in Alexandr Wang and rebuild the AI organization from scratch.
According to Meta, the team "rebuilt its AI stack from the ground up" over the past nine months — new training techniques, new infrastructure, and a new scaling approach where each generation validates before going bigger. Muse Spark is deliberately "small and fast by design" — a foundation for larger future models already in development.
This is the most controversial part of the announcement. Meta built its developer credibility on open-source Llama models — a strategy that attracted hundreds of thousands of developers and positioned Meta as the "Linux of AI." Muse Spark breaks from that tradition.
Meta says it "hopes to open-source future versions," and Wang has indicated that some models in the Muse series will remain proprietary while others may be released publicly. The shift reflects a more pragmatic hybrid strategy: keep the most capable models closed, release smaller or older versions for the community.
For developers who relied on Llama, this is a meaningful change. The Llama ecosystem isn't going away, but its successor is no longer guaranteed to be open by default.
WhatsApp users: The biggest near-term impact. Hundreds of millions of people in countries where WhatsApp is the primary communication platform will get access to a significantly smarter AI assistant inside the app they already use every day — no subscription required.
Businesses on Meta's platforms: Shopping mode and the creator-tied recommendations create new advertising and commerce surfaces. If you run a brand or sell products via Instagram or Facebook, expect new AI-powered discovery features to become relevant to your strategy.
Enterprise developers: Currently limited to private API preview. If you're building on Meta's platforms, watch for the API to open up — Muse Spark's multimodal and multi-agent capabilities could be useful for social commerce and customer service applications.
Researchers and power users: Muse Spark's health reasoning capabilities are notable. It scores well on medical benchmarks, which could make it a useful tool for health-adjacent applications — though as always, it should never replace professional medical advice.
Muse Spark is a genuine comeback for Meta AI — not the industry leader, but no longer an afterthought. It sits just behind GPT-5.4, Gemini 3.1 Pro, and Claude Opus 4.6 on pure intelligence benchmarks, while offering unique strengths in multimodal perception and health reasoning. Its real power comes from distribution: 3 billion users, zero subscription required. For most consumers, the best AI isn't necessarily the smartest one — it's the one that's already in their pocket. Muse Spark is about to be in a lot of pockets.
Meta is back in the AI race. Muse Spark isn't #1 yet, but it's competitive, broadly distributed, and the foundation of a larger roadmap. Watch for the WhatsApp rollout — that's when the real-world impact becomes clear.
We've added Meta AI (Muse Spark) to our tool directory. You can compare it against other leading AI assistants on our Compare page.
Take our 5-question quiz and get a personalized recommendation — including whether Muse Spark fits your needs.
🎯 Find My AI →