Meta’s new AI model is giving its competitors a scare Everything you need to know about Muse Spark
Recently Meta introduced its first generative artificial intelligence model. It is called Muse Spark and will soon be integrated into WhatsApp, Instagram, and the “smart” glasses produced by the company. To achieve this result, the company led by Mark Zuckerberg chose to scale back investments in the Metaverse, the immersive technology it had previously strongly focused on (even going so far as to change its name), but which never truly managed to take off.
Muse Spark was developed by the Superintelligence Lab, a division of the company led by Alexandr Wang, former CEO of Scale AI, a company that Meta largely acquired in 2025 with an investment of over 14 billion dollars.
How Meta managed to develop Muse Spark
Introducing Muse Spark, the first in the Muse family of models developed by Meta Superintelligence Labs.
— AI at Meta (@AIatMeta) April 8, 2026
Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration.
Muse Spark is available today at… pic.twitter.com/qnfSzoSPzt
Muse Spark was highly anticipated because Meta had lagged behind in the AI field in recent years compared to other big tech companies. For this reason, Zuckerberg decided to overhaul the company’s artificial intelligence division in April 2025, following the disappointing results achieved by the previous model Llama 4.
The company also allocated tens of billions of dollars to a massive hiring campaign, with the goal of catching up on the delay accumulated over the years. This strategy allowed it to attract — and often poach from competitors — the most sought-after developers and researchers in the artificial intelligence sector, thanks to compensation packages that in some cases reached 100 million dollars per year.
What people like and dislike about Muse Spark
@rpn Meta just released Muse Spark, which will change your experience on Instagram, Facebook, WhatsApp, etc. Do you remember the all-star team Zuck recruited to form Meta Superintelligence Labs? This is their very first release — which has been highly anticipated. Now it isn’t state-of-the-art right out of the gates, but it is extremely competitive with some of today’s top models like Opus 4.6 and GPT 5.4 For example: On SWE-Bench Verified, Muse Spark scored a 77.4. Opus 4.6 is at 80.8, Gemini 3.1 Pro is at 80.6 and Grok 4.2 is at 76.7. On Humanity’s Last Exam, Muse Spark scored a 42.8, compared to Opus 4.6’s 40.0 and GPT 5.4’s 52.1. That’s impressive for a 9-month old lab. One of the things I love about the AI race, is how hands-on founder have been. Sergey Brin is back to continue the mission at Google. Tobi Lutke is coding again at Shopify. And @zuck has been personally involved, from what I hear — spending an enormous amount of time in the trenches with the engineers and researchers at MSL. Creators and entrepreneurs should be paying attention to major opportunity the Muse family of models will unlock for commerce. The bet is essentially that whoever owns the AI layer between a person and their purchasing decisions owns commerce at scale. And the personal context advantage Meta has is massive. This model is just the beginning and may be the final piece that was missing. If you like this video, follow @rpn or comment “signal” for my newsletter to stay two steps ahead on the incredibly disruption happening in tech and how to leverage it in content and business. #meta #msl #spark #muse #ai original sound - Roberto Nickson
Most companies involved in the development of AI systems are focusing their efforts — and the largest share of their investments — on software development. In this area, however, Muse Spark’s performance still falls short compared to its main competitors. In terms of text generation, on the other hand, the tool appears capable of reaching levels comparable to those of ChatGPT, Gemini, or Claude.
Meta has nevertheless announced that it is already working on a new artificial intelligence model, known in the industry by the codename “Watermelon”, which is expected to offer superior performance in code generation. In light of this development as well, Zuckerberg has indicated that in the coming years a significant share of the company’s resources will be allocated to AI. In particular, Meta plans to invest an additional 600 billion dollars in building new data centers, which are essential for the operation of artificial intelligence systems.