After months of false starts, Meta returns with an AI model capable of standing up to its rivals. Muse Spark is not just a technical advance, it is the sign of a profound strategic shift. This launch says a lot about Meta's ambitions. And perhaps even more about the end of an era.
In brief
- Meta launched Muse Spark on Wednesday, its first major AI model in more than a year.
- The announcement propelled META shares up to +9% during the session.
- Alexandr Wang, new AI director of Meta, is piloting this overhaul from the Superintelligence Laboratories.
Muse Spark, the new face of AI at Meta
Meta formalized on Wednesday the launch of Muse Spark, its new AI model, now available via its dedicated site and application. This launch marks the strong return of the Menlo Park firm after more than a year without much news on the AI front.
At the helm of this project: Alexandr Wang, 28 years old, founder of Scale AI and new AI director of Meta. Wang joined the company after Meta acquired 49% of Scale AI for about $14 billion. He now oversees Superintelligence Laboratories, the division responsible for pushing the group's AI ambitions to their maximum.
The markets welcomed the announcement: META stock jumped as much as 9%suddenly erasing the losses accumulated since the end of March. A strong signal, which shows that investors still believe in Zuckerberg's ability to turn things around after recent disappointments.
Performance-wise, Meta claims that Muse Spark performs the same tasks as its predecessor Llama 4 Maverick, but with significantly reduced computing consumption.
Internal benchmarks show that the model equals, or even exceeds on certain criteria, solutions from Google, OpenAI and Anthropic. Independent tests confirm solid results in the medical field, but still point to programming gaps.
A strategic turning point that does not speak its name
This launch is much more than a technical update. Muse Spark breaks with one of Meta's most entrenched traditions: open source. Unlike the Llama series, this new model is closed and proprietary.
Some partners will be able to access it via an API in preview, but Meta has not yet committed to a public release. The company is talking about future versions that are potentially open source, but without a precise timetable.
This turnaround illustrates the growing pressure Meta is under in the AI race. The Llama 4 model disappointed when it was launched in 2025. The Behemoth project, a behemoth of 2,000 billion parameters, was abandoned before even seeing the light of day. Faced with these setbacks, management has decided: we must move quickly, and protect what can still be protected.
To finance this race, Meta is not skimping: $135 billion in AI investments are planned for 2026, almost double the previous year. A colossal sum, which does not come without social consequences.
The Reality Labs division, that of the Metaverse, which swallowed up $80 billion, suffered hundreds of layoffs. And according to several sources, new workforce reduction plans are being studied, potentially exceeding 20% of employees. Meta denies, but doubt sets in.
This context also sheds light on information that has passed in the background: Zuckerberg himself is developing a personal AI agent to control Meta with fewer intermediaries. AI as an internal management tool, as a hierarchical shortcut, is perhaps the true face of this revolution.
Muse Spark is just the beginning, Meta executives warn. More advanced models are announced for this year. The AI race is on again.
Maximize your Tremplin.io experience with our 'Read to Earn' program! For every article you read, earn points and access exclusive rewards. Sign up now and start earning benefits.
