Details
- Meta announced Muse Spark on April 8, 2026, the inaugural model in its new Muse series, designed as a small, fast foundation for scaling toward personal superintelligence with complex reasoning in science, math, and health.
- Meta Superintelligence Labs, led by former Scale AI CEO Alexandr Wang, rebuilt Meta's AI stack from the ground up over nine months after Mark Zuckerberg's dissatisfaction with prior Llama models lagging OpenAI and Anthropic.
- Muse Spark powers upgraded Meta AI with multimodal perception, multi-agent orchestration (e.g., parallel subagents for trip planning), visual coding for websites/games, shopping mode from Instagram/Facebook content, and physician-tuned health responses including images/charts.
- Initially available on Meta AI app and meta.ai in the US with Instant/Thinking modes; rolling out to WhatsApp, Instagram, Facebook, Messenger, and AI glasses in coming weeks, plus private API preview for partners and planned open-sourcing of future versions.
- Early benchmarks show Muse Spark dominating health tasks (e.g., Health Bench Heart at 42.8% vs. Gemini's 20%, Opus's 14%), natively multimodal with tool-use and visual chain-of-thought, reaching Meta's 3B+ daily users across apps.
Impact
Muse Spark accelerates Meta's AI competitiveness against OpenAI, Anthropic, and Google by leveraging its 3B+ user base for rapid iteration and multimodal personalization, potentially boosting adoption in health/consumer AI where it leads benchmarks. Over 12-24 months, expect intensified R&D in multi-agent systems and on-device inference via AI glasses, driving funding toward social-graph integrated superintelligence while raising privacy scrutiny on user data training.
