Meta's Superintelligence Labs pushes beyond lightweight Muse Spark with a larger, higher-benchmark model aimed at coding and complex reasoning tasks.
Meta released Muse Wonder, positioning it as the company's first true frontier-class model after the lightweight, low-latency Muse Spark. The new models post higher benchmark scores and are built for harder workloads like coding, according to the company.
The launch follows Meta's earlier debut of the Muse agent and Muse Spark, signaling an accelerated cadence from the reorganized Superintelligence Labs division. Meta is explicitly trying to close the gap with OpenAI, Google, and Anthropic on capability, not just efficiency.
Meta has spent two years playing catch-up on frontier capability while leading on open distribution. A credible frontier model changes Meta's enterprise pitch from 'free and fast' to 'free and competitive,' which pressures pricing across the model market that every AI-dependent business buys into.
The daily signal, curated. Get it in your inbox.
Subscribe on LinkedIn →