Meta has introduced Muse Spark, a new AI model designed to scale predictably and efficiently toward personal superintelligence. The company has focused on refining the pretraining process, which provides the model with its core multimodal understanding, reasoning, and coding skills. Over the past nine months, Meta's research and engineering teams have overhauled the pretraining stack by improving model architecture, optimization procedures, and data selection.
1/ today we're releasing muse spark, the first model from MSL. nine months ago we rebuilt our ai stack from scratch. new infrastructure, new architecture, new data pipelines. muse spark is the result of that work, and now it powers meta ai. 🧵 pic.twitter.com/fThDXdsxwB
— Alexandr Wang (@alexandr_wang) April 8, 2026
These changes allow Muse Spark to achieve the same performance as earlier iterations, such as Llama 4 Maverick, but with more than ten times less computational power required.
Meta jumped from last to the 4th place on Artificial Analysis arena with its newly released Muse Spark model.
— TestingCatalog News 🗞 (@testingcatalog) April 8, 2026
It also appears to be token efficient for its level of intelligence. https://t.co/uozTYUcEXK pic.twitter.com/ATH9r6YF18
The new model is currently being evaluated through rigorous benchmarking, where its training efficiency and performance have been compared across a range of smaller models using specific scaling laws.
💠Meta has launched a new visual identity for Meta AI that includes a brand new logo and a new colour for the app’s main buttons.@testingcatalog pic.twitter.com/s7Pns0SoXu
— Radu Oncescu (@oncescuradu) April 8, 2026
Meta continues to push boundaries in the AI field with a focus on efficiency and scalability, aiming to provide tools that can support broader applications in both research and commercial settings. The release is currently available to select partners and research teams, with broader access expected as the technology matures and further validation is completed.
