Meta has launched Muse Spark, a new AI model created by Meta Superintelligence Labs. The model belongs to the Muse series, which focuses on understanding different types of content, reasoning, and working with AI agents.
Meta states that Musk Spark is a key component of a larger revamp of its AI ecosystem, encompassing model development, infrastructure, and deployment processes. The model is now available via Meta AI and the Meta AI app, with a private API preview for select users.
1/ today we're releasing muse spark, the first model from MSL. nine months ago we rebuilt our ai stack from scratch. new infrastructure, new architecture, new data pipelines. muse spark is the result of that work, and now it powers meta ai. 🧵 pic.twitter.com/fThDXdsxwB
— Alexandr Wang (@alexandr_wang) April 8, 2026
What Muse Spark does:
Muse Spark works as a foundation model that handles both text and images, performs structured reasoning, uses tools, and enables multiple AI agents to collaborate.
Main features:
Works with text and image inputs
Built-in reasoning abilities
Can use tools
Processes visual chain-of-thought
Runs multiple agents at once
The model is compact and quick while managing difficult tasks in fields like science, math, and health. It’s part of a larger update to Meta’s AI infrastructure, backed by the Hyperion data center.
Performance against competitors
Meta says Muse Spark beats rival models including Gemini 3.1 Pro, GPT-5.4, and Grok 4.2 in multiple tests.
In CharXiv reasoning tests, Muse Spark got 86.4 points, beating GPT-5.4 (82.8), Gemini 3.1 Pro (80.2), and Grok 4.2 (60.9). For SimpleVQA, it scored 71.3. In ZeroBench, it reached 33.0, topping Gemini’s 29.0 but falling short of GPT-5.4’s 41.0.
For health tests, Muse Spark achieved 42.8 in HealthBench Hard, surpassing Gemini 3.1 Pro (20.6) and Grok 4.2 (20.3), and edging past GPT-5.4 (40.1).
But Meta’s own numbers show Muse Spark trails behind in some tests. Gemini and GPT models scored higher in MMMU Pro and ARC AGI 2, showing were Muse Spark needs improvement.
Safety and access
Meta tested Muse Spark for safety using its Advanced AI Scaling Framework. The model refuses requests in high-risk areas and doesn’t show harmful autonomous behavior.
You can start using Muse Spark today through Meta AI platforms. Features like Contemplating mode will become available over time.
The arrival of Muse Spark shows just how fierce the race has become between tech giants trying to create the best AI helpers. Meta, OpenAI, Anthropic, and Google are all pouring massive amounts of money into AI development and the technology needed to support it. But Meta’s taking a unique approach—they want to weave AI throughout their social media apps, where people are already spending time by the billions.Sonnet 4.5