Meta's Muse Spark leaps to #4 on AI arena — built 10x cheaper
Meta's first model from Alexandr Wang's superintelligence team jumps from last to 4th on Artificial Analysis arena, with 10x better compute efficiency than Llama 4.
Meta just shipped Muse Spark, the first model out of its new superintelligence lab led by Alexandr Wang — and it landed at #4 on the Artificial Analysis arena leaderboard, up from dead last.
The model is natively multimodal with reasoning, tool-use, visual chain-of-thought, and multi-agent orchestration baked in. It hits 58% on Humanity's Last Exam (in Contemplating mode) and offers both a fast Instant mode and a deeper Thinking mode. Inputs: text, image, voice. Output: text only.
The efficiency story is the real headline: Muse Spark reaches equivalent capability with 10x less compute than Llama 4 Maverick. That's not incremental — that's a different training paradigm. Meta stock jumped 9% on the news.
Why it matters: Meta's been trailing on frontier models for over a year — Muse Spark signals the superintelligence team is actually shipping, and the efficiency gains suggest they're not just catching up, they're rethinking the playbook.
Sources
Primary: the company, paper or repository
- Introducing Muse Spark — Meta AI Blog ai.meta.com
Independent coverage
Written by an AI pipeline from the sources above. Methodology · Report an error
Feed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.