Meta's Superintelligence Lab Debuts Spark, a Proprietary Break from Llama
Original source
Meta's Superintelligence Lab unveils its first public model, Muse Spark
Ars Technica →Meta has released Muse Spark, the first model from its Superintelligence Labs division - a unit formed roughly a year ago with ambitions around personal superintelligence. The release marks a deliberate departure from the Llama open-source lineage, which has underperformed on user satisfaction and independent benchmarks. Spark is proprietary, though Meta has signaled future open-source releases under the Muse umbrella.
Spark is deeply integrated with Meta’s social platforms, drawing on content from Instagram, Facebook, and Threads in a manner analogous to how xAI’s Grok leverages X. Initial capabilities include surfacing relevant public posts by location or trending topic; planned expansions include recommendation citations and inline Reels and photos with creator attribution.
On benchmarks, Meta claims Spark’s standard reasoning mode is competitive with or superior to models from OpenAI, Anthropic, Google, and xAI. The company simultaneously acknowledges meaningful gaps in long-horizon agentic tasks and coding workflows - two areas that have become primary evaluation criteria as the model race shifts from raw language fluency toward autonomous task execution.
Read the full article
Continue reading at Ars Technica →This is an AI-generated summary. Read the original for the full story.