Muse Spark — Meta's Most Powerful AI Model

What Is Meta Muse Spark?

Meta Muse Spark is a frontier AI model released on April 8, 2026 by Meta Superintelligence Labs (MSL), a new research division within Meta led by Alexandr Wang, former CEO of Scale AI. Muse Spark represents a complete departure from Meta's Llama model family — it is not Llama 5, but an entirely new model architecture designed to compete at the very top of the frontier AI landscape.

Unlike the open-weight Llama models that preceded it, Muse Spark is closed source — a significant strategic shift for Meta, which built its AI reputation on open-source releases. The model is currently available for free at meta.ai, while API access is in private preview for select partners.

The Muse Spark benchmark results upon release were striking: it claimed the #1 position on CharXiv (chart understanding), HealthBench Hard (medical reasoning), FrontierScience (scientific frontier reasoning), and HLE in Contemplating Mode (humanity's last exam). These results placed it ahead of GPT-5.4 and Gemini 3.1 Pro on most scientific and medical tasks — though it trails on ARC AGI 2 and Terminal-Bench 2.0.

Muse Spark Contemplating Mode Explained

The most distinctive feature of Muse Spark is its Contemplating Mode — a multi-agent parallel reasoning system that spawns and coordinates multiple AI agents simultaneously to solve a single complex problem. This is distinct from simple chain-of-thought reasoning: rather than a single model reasoning step-by-step, Contemplating Mode runs many parallel reasoning threads, synthesizes their outputs, and produces a final answer informed by the ensemble.

Contemplating Mode is what enables Muse Spark's HLE score of 50.2 — the highest recorded on Humanity's Last Exam, a benchmark designed to be unsolvable by current AI systems. The mode is best suited for tasks requiring exhaustive reasoning: graduate-level science problems, complex mathematical proofs, multi-step medical diagnoses, and frontier research analysis.

For everyday tasks, Muse Spark's Fast Mode provides quick, high-quality responses without the computational overhead of multi-agent coordination. Users can switch between modes depending on whether they need speed or depth.

Meta Superintelligence Labs & Alexandr Wang

Meta Superintelligence Labs (MSL) was established in early 2026 as a dedicated AI research division within Meta, separate from the FAIR (Fundamental AI Research) team that produced the Llama series. MSL was created with an explicit mandate: build the world's most capable AI model, without the constraints of open-source release timelines.

Alexandr Wang, who built Scale AI into the leading AI data infrastructure company before joining Meta, leads MSL. Wang's background in high-quality AI training data is widely credited as a key factor in Muse Spark's performance on scientific and medical benchmarks — domains where data quality and curation are critical.

The decision to make Muse Spark closed source — while keeping meta.ai free — reflects a new strategy: use frontier model capability to drive engagement with Meta's consumer products while reserving commercial API revenue for business use cases. This puts Meta in more direct competition with OpenAI and Google DeepMind on the enterprise AI platform market.

Muse Spark vs GPT-5: How Do They Compare?

The Muse Spark vs GPT-5 comparison is the most-searched question since the model's release. The short answer: Muse Spark leads on scientific and medical tasks, while GPT-5.4 leads on overall arena scores and agentic coding benchmarks.

On CharXiv (chart understanding and reasoning), Muse Spark scores 86.4 vs GPT-5.4's 82.8 — a meaningful gap in a benchmark that tests visual and analytical reasoning together. On HealthBench Hard, Muse Spark's 42.8 significantly outperforms GPT-5.4's 40.1 and Gemini 3.1 Pro's 20.6, suggesting a particular strength in structured medical reasoning.

However, on ARC AGI 2 (novel problem-solving that resists training data memorization), Muse Spark scores 42.5 versus GPT-5.4's 76.1 and Gemini 3.1 Pro's 76.5 — a substantial gap. This suggests that while Muse Spark excels at knowledge-intensive tasks, it currently lags competitors on tasks requiring generalization to truly novel problems.

For most professional use cases — research assistance, medical information, scientific literature analysis, and complex document understanding — Muse Spark's 262K context window and multimodal capabilities make it a compelling choice, especially given its free availability at meta.ai.

Muse Spark API: When Will It Be Available?

As of April 2026, the Muse Spark API is in private preview for select Meta partners. No public release date has been announced. Meta has stated that API access will be expanded in phases, with priority given to healthcare, education, and enterprise research partners.

The Muse Spark API is expected to support multimodal input (text, image, and voice), both Fast Mode and Contemplating Mode, and context windows up to 262K tokens. Pricing details have not been released, but given Meta's stated goal of competing with OpenAI on enterprise AI, pricing is expected to be competitive with GPT-5.4's API rates.

Subscribe below to be notified when the Muse Spark API opens for public access. In the meantime, you can explore the full model for free at meta.ai, which provides access to both Fast Mode and Contemplating Mode without any usage limits at launch.