Meta unveils Muse Spark, its first superintelligence model to rival OpenAI and Google

Meta is stepping back into the AI spotlight with a new model it hopes will reset the narrative. On April 8, Meta Platforms introduced Muse Spark, the first release from its newly formed Meta Superintelligence Labs. The model marks a turning point for the company after a year of heavy spending, internal restructuring, and a push to close the gap with leaders like OpenAI and Google.

The stakes are high. Meta has poured tens of billions into rebuilding its AI efforts after its Llama 4 models failed to impress earlier in the cycle. Bringing in Alexandr Wang to lead the new division was a clear signal that the company was willing to rethink its entire approach.

Muse Spark is the first real output of that reset.

The model handles text, images, and voice, and introduces a “contemplating mode” that runs multiple agents simultaneously to tackle harder problems. In early benchmarks, it holds its own against top-tier models on reasoning and multimodal tasks, though it still trails in areas like coding and abstract problem-solving.

Inside Meta’s Muse Spark: A New AI Model Signals a Strategic Shift Away from Llama

Meta describes the system as “small and fast by design,” built to handle complex questions across science, math, and health. The company is putting extra weight behind health-related reasoning as part of its broader push into what it calls personal superintelligence.

“This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math, and health. It is a powerful foundation, and the next generation is already in development,” the company said in a blog post.

Meta Muse Spark (Courtesy: Meta)

The release carries another major shift. Unlike the Llama family, Muse Spark is not open. Meta is limiting access to a private preview with select partners, a sharp departure from its earlier open-source stance. Future versions could be shared more broadly, though there is no timeline.

Earlier this year, pressure inside Meta’s AI division was already spilling into public view. Low-cost models from DeepSeek were gaining traction and outperforming rivals on efficiency, raising fresh questions about Meta’s strategy. In January, TechStartups reported on an anonymous post on Blind forum that described the company’s generative AI team as being in “panic mode.”

That decision points to a larger change in strategy. Meta appears focused on protecting its latest work, keeping tighter control as competition intensifies and model development costs continue to climb.

Internally, Muse Spark is part of a new model line codenamed Avocado. It is already being integrated into Meta’s ecosystem, starting with the Meta AI app and website, with plans to expand across WhatsApp, Instagram, Facebook, and the company’s smart glasses in the coming weeks.

That distribution advantage may matter as much as the model itself. Meta can deploy updates across more than 3.5 billion users, giving it a testing ground few competitors can match. The company is already hinting at how it plans to turn that reach into revenue, including shopping features inside its chatbot that guide users directly to products.

Early independent evaluations place Muse Spark in the top tier across a broad range of tests, with strong performance in language and visual tasks. It still sits behind the very best models in some categories, tying for fourth place on an index compiled by Artificial Analysis.

Meta executives are not overselling the first release. Mark Zuckerberg told investors earlier this year that the initial models would show progress over perfection, pointing to a steady pace of improvement over time.

Wang echoed that sentiment after the launch, writing that “there are certainly rough edges we will polish over time in model behavior.” Larger versions of Muse Spark are already in development, and Meta plans to release more models from the Avocado line over the coming year.

The bigger story is what Muse Spark represents.

After years of pushing open models, Meta is shifting toward a more controlled approach. At the same time, it is leaning into its strongest asset: distribution. By embedding AI directly into products people already use every day, the company is betting that engagement will grow faster than its rivals can match.

Meta is not leading the field yet. The gap with top models from OpenAI and Google remains in key areas. Still, the direction has changed. Muse Spark signals a company that is back in the race, with a clearer strategy and the resources to keep pushing forward.