Meta Platforms announced a new large language model Wednesday, its first major new artificial-intelligence model in more than a year.
The rollout of the model, called Muse Spark, is a critical moment for Meta, which has spent billions of dollars hiring AI talent in a bid to catch up to OpenAI, Anthropic and Google DeepMind. The leading labs have been putting out models at an accelerating pace.
In a departure from its previous models, which were open-source, Muse Spark is a closed model that will power Meta's AI chatbot and AI features within it. Meta's longer-term goal is to build what it calls superintelligence, smarter-than-humans technology that would power personal agents able to handle tasks for all of the company's more than one billion users.
The company said it planned to release a private preview of the model to a few partners via an application programming interface, or API, which allows developers to build on top of existing software, and at some later point might open-source some versions of the model. Open-sourcing a model gives the public access to some parts of the code and other architecture behind it.
According to Meta's internal benchmark tests, Muse Spark outscored Google's Gemini on some tests and was competitive with models from OpenAI and Anthropic on others. It significantly outscored xAI's Grok on most tests. The company said the addition of Muse Spark has made Meta AI more effective at answering health-related questions, one of the top consumer uses of AI.
"Meta just did a step change from Llama 4 to this," said Rayan Krishnan, chief executive of Vals AI, an independent startup that does testing of new frontier models and tested Muse Spark ahead of its public announcement; Llama 4 was Meta's previous model. "They're now a competitive lab. If the rate of progress stays, it's not hard to imagine them producing a state-of-the-art model in a short period of time."
However, Krishnan said, "the model is still underperforming on coding, so I would expect that to be a domain where they double down in the future."
Meta Chief Executive Mark Zuckerberg sough to manage expectations for the model on an earnings call in January.
"I expect our first models will be good but more importantly will show the rapid trajectory that we're on, and then I expect us to steadily push the frontier over the course of the year as we continue to release new models," he said.
He echoed that sentiment Wednesday, saying "we plan to release increasingly advanced models that push the frontier of intelligence and capabilities, including new open source models."
Meta's stock rose roughly 4% after the model was announced and ended trading up 6.5%.
Colin Sebastian, a senior research analyst at Baird, called the model announcement an important milestone on a longer road to "better monetization and more user engagement."
"The next steps are certainly also important," he said. "Do we see improvement in the core apps? Are they able to convince users to use Meta AI for things that they're using ChatGPT and Gemini for today?"
The announcement Wednesday represents an achievement for Alexandr Wang, the former Scale AI CEO whom Zuckerberg hired as part of a $14 billion deal and put to work overseeing Meta's new AI efforts last summer.
Wang's hiring followed the disappointing release of Llama 4. The company was accused of -- and later admitted to -- gaming a third-party benchmark that is used to rank various models against each other on performance. It also delayed the rollout of its biggest model, called Behemoth, which it never ultimately released.
In his bid to catch up to rivals, Zuckerberg doled out $100 million offers to AI researchers and wooed them at his homes in Palo Alto, Calif., and Lake Tahoe. He eventually assembled a team of 50-plus researchers, engineers and other AI employees to be part of the new effort he dubbed Meta Superintelligence Labs.
The influx of new hires -- including the appointment of a new chief scientist -- created friction with some earlier AI employees. Many left the company, either voluntarily for rivals or via layoffs that came in October as part of the restructuring.
In December, Wang disclosed during an internal company Q&A that his team was working on two new models, a text-based LLM code-named Avocado and an image-and-video-focused model code-named Mango.
