Meta Launches New AI Model Muse Spark, Aiming at OpenAI
Meta has announced its new AI model, Muse Spark, as the starting point for its new Muse series, directly challenging OpenAI. This model is designed to be small and fast, capable of reasoning complex problems in science, mathematics, and health. Meta also introduced 'Contemplating Mode' to enhance reasoning by running multiple agents simultaneously, competing with Google's Gemini Deep Think and OpenAI's GPT Pro. Muse Spark will initially be available in the Meta AI app and will eventually power AI features across WhatsApp, Facebook, and Meta smart glasses.
📋 Article Processing Timeline
- 📰 Published: April 9, 2026 at 11:07
- 🔍 Collected: April 9, 2026 at 12:00 (53 min after Published)
- 🤖 AI Analyzed: April 15, 2026 at 18:04 (150h 4m after Collected)
Reuters reported that Meta stated it has rebuilt its artificial intelligence (AI) technical architecture from scratch over the past nine months, with Muse Spark serving as the starting point for the new Muse series.
U.S. tech giants face pressure to prove that massive AI spending will pay off. For Meta, this gamble is particularly significant, as the company hired Scale AI CEO Wang Tao with a high salary last year and offered some engineers hundreds of millions of dollars in compensation packages to build a new super-intelligence team, striving to re-enter the top tier of AI after the disappointing performance of the Llama 4 model early last year.
Meta mentioned Muse Spark in a blog post: "This initial model is designed to be small and fast, yet capable enough to reason complex problems in science, mathematics, and health. It is a powerful foundation, and the next generation of products is already under development."
In addition, Meta also launched "Contemplating Mode," which enhances reasoning capabilities by simultaneously executing multiple agents, allowing Muse Spark to compete with deep thinking modes like Google's Gemini Deep Think and OpenAI's GPT Pro.
In practical applications, Muse Spark can help users estimate food calories, perform image simulations, and even plan family travel itineraries.
Independent evaluations show that the model is close to market leaders (such as Google, OpenAI, Anthropic) in language and visual understanding, but still has a gap in programming and abstract reasoning, ranking fourth overall.
The new model is currently only available in the United States. Unlike its previous strategy of prioritizing open-source models, Muse Spark is initially only available in the Meta AI application and website, and is offered as a "private preview" to some partners, indicating a shift from its past open-source strategy to a more closed model.
Muse Spark will initially be available in the Meta AI application, but will soon fully drive various AI functions in WhatsApp, Facebook, and Meta smart glasses, reaching over 3.5 billion users.
In terms of business model, Meta is exploring integrating AI with e-commerce, for example, embedding shopping functions in chatbots to guide users to directly purchase products. (Compiler: Cheng Shih-yun) 1150409
U.S. tech giants face pressure to prove that massive AI spending will pay off. For Meta, this gamble is particularly significant, as the company hired Scale AI CEO Wang Tao with a high salary last year and offered some engineers hundreds of millions of dollars in compensation packages to build a new super-intelligence team, striving to re-enter the top tier of AI after the disappointing performance of the Llama 4 model early last year.
Meta mentioned Muse Spark in a blog post: "This initial model is designed to be small and fast, yet capable enough to reason complex problems in science, mathematics, and health. It is a powerful foundation, and the next generation of products is already under development."
In addition, Meta also launched "Contemplating Mode," which enhances reasoning capabilities by simultaneously executing multiple agents, allowing Muse Spark to compete with deep thinking modes like Google's Gemini Deep Think and OpenAI's GPT Pro.
In practical applications, Muse Spark can help users estimate food calories, perform image simulations, and even plan family travel itineraries.
Independent evaluations show that the model is close to market leaders (such as Google, OpenAI, Anthropic) in language and visual understanding, but still has a gap in programming and abstract reasoning, ranking fourth overall.
The new model is currently only available in the United States. Unlike its previous strategy of prioritizing open-source models, Muse Spark is initially only available in the Meta AI application and website, and is offered as a "private preview" to some partners, indicating a shift from its past open-source strategy to a more closed model.
Muse Spark will initially be available in the Meta AI application, but will soon fully drive various AI functions in WhatsApp, Facebook, and Meta smart glasses, reaching over 3.5 billion users.
In terms of business model, Meta is exploring integrating AI with e-commerce, for example, embedding shopping functions in chatbots to guide users to directly purchase products. (Compiler: Cheng Shih-yun) 1150409