Meta’s Muse Spark shows up on leaderboard
Meta introduced Muse Spark — its first proprietary model outside the open‑weight Llama line — and Muse Spark placed #3 for text and #2 for vision in the LMSYS Arena, tying with Claude Opus‑4.6 in that ranking (x.com). The post framed Muse Spark as Meta’s move toward closed, proprietary models rather than continuing purely with Llama‑style open weights (x.com).
Meta’s new Muse Spark model has climbed to No. 3 in Arena’s text rankings and No. 2 in vision, days after Meta unveiled it on April 8. (arena.ai) (about.fb.com) Arena’s leaderboard showed Muse Spark at 1,493 in text, behind Anthropic’s Claude Opus 4.6 Thinking at 1,504 and Claude Opus 4.6 at 1,496, based on thousands of user votes. In vision, Muse Spark scored 1,293, ahead of Claude Opus 4.6 at 1,289 and behind only Claude Opus 4.6 Thinking at 1,302. (arena.ai) Meta said Muse Spark is the first model in a new Muse series built by Meta Superintelligence Labs, and the company said it already powers the Meta AI app and website. Meta also said the model will roll out to WhatsApp, Instagram, Facebook, Messenger, and its artificial intelligence glasses in the coming weeks, with a private-preview application programming interface for select partners. (about.fb.com) The release marks a break from Meta’s recent pattern with Llama, the company’s open-weight model line that developers could download and run under Meta’s license. Artificial Analysis said Muse Spark is Meta’s first release since Llama 4 in April 2025 and Meta’s first release that is not open weights. (artificialanalysis.ai) (business-standard.com) Meta tied Muse Spark to the artificial intelligence reorganization it started in mid-2025. Meta executive bio pages say Alexandr Wang joined in June 2025 as the company’s first chief artificial intelligence officer, and CNBC reported on June 30, 2025 that Mark Zuckerberg created Meta Superintelligence Labs under Wang’s leadership. (meta.com) (cnbc.com) Meta described Muse Spark as “small and fast by design” and said it was built for reasoning and multimodal work, meaning it can handle images as well as text. The company said the model can answer complex questions in science, math, and health, and can inspect photos, products, and scenes for Meta AI users. (about.fb.com) CNBC reported that Meta is not pitching Muse Spark as its largest model, but as an efficient one, and said the company plans to test a new revenue stream by eventually offering outside developers access through an application programming interface. The same report said Meta was trying to regain ground against OpenAI, Anthropic, and Google after a weak reception for its previous flagship models. (cnbc.com) For now, Muse Spark’s leaderboard showing gives Meta something it has lacked in recent months: a fresh model near the top of a public ranking that compares systems side by side. Meta said larger Muse models are already in development. (arena.ai) (about.fb.com)
