Subscribe
Share
Share
Embed
Meta has launched Muse Spark, the first artificial intelligence model developed by its dedicated superintelligence team. Led by Alex Wang, this release marks the beginning of the internal "Avocado" series, designed to help Meta regain its standing following the underwhelming performance of its Llama 4 models last year. While Muse Spark is competitive with leaders like OpenAI and Google in language and visual understanding, it currently lags in coding and abstract reasoning tasks. The model is being introduced via a private preview and features a new "Contemplating Mode" that utilizes multiple agents to solve complex problems like travel planning. This launch also signals a shift in Meta’s strategy, moving away from its traditional open-source approach to prioritize integration within its social platforms and new shopping features.
Meta has officially debuted Muse Spark, the inaugural model from the high-stakes superintelligence team assembled last year under the leadership of Alex Wang. Reported by The Guardian, this release represents a critical test for the company following a $14.3 billion investment to recruit top-tier talent and catch up with rivals like OpenAI and Google. Muse Spark is the first entry in the internal "Avocado" series and is designed to be small and fast while handling complex reasoning in health, math, and science. However, benchmarks from Artificial Analysis indicate that while the model is strong in language, it still struggles with coding and abstract reasoning. Meta is also pivoting its distribution strategy, opting for a private preview rather than the open releases seen with previous Llama iterations, as it seeks to monetize the technology through embedded shopping features and advanced multi-agent capabilities.
Neural Newscast is AI-assisted, human reviewed. View our AI Transparency Policy at NeuralNewscast.com.
Neural Newscast delivers clear, concise daily news - powered by AI and reviewed by humans. In a world where news never stops, we help you stay informed without the overwhelm.
Our AI correspondents cover the day’s most important headlines across politics, technology, business, culture, science, and cybersecurity - designed for listening on the go. Whether you’re commuting, working out, or catching up between meetings, Neural Newscast keeps you up to date in minutes.
The network also features specialty shows including Prime Cyber Insights, Stereo Current, Nerfed.AI, and Buzz, exploring cybersecurity, music and culture, gaming and AI, and internet trends.
Every episode is produced and reviewed by founder Chad Thompson, combining advanced AI systems with human editorial oversight to ensure accuracy, clarity, and responsible reporting.
Learn more at neuralnewscast.com.
[00:00] Announcer: From Neural Newscast, this is Model Behavior, AI-focused news and analysis on the models shaping our world.
[00:11] Nina Park: I'm Nina Park. Welcome to Model Behavior.
[00:15] Nina Park: Today, we examine how Meta is attempting to reclaim its position in the artificial intelligence landscape with a new model release.
[00:24] Thatcher Collins: And I'm Thatcher Collins.
[00:25] Thatcher Collins: Yesterday, Meta unveiled MuseSpark, the first product from the super intelligence team they spent billions to assemble.
[00:33] Thatcher Collins: It marks a significant moment, Nina, as it follows a very expensive and very public effort to correct the trajectory they were on after Lama 4 struggled last year.
[00:44] Nina Park: The scale of that investment is hard to overstate.
[00:47] Nina Park: As reported by The Guardian, Meta spent over $14 billion to recruit Alex Wang and a team of engineers, some of whom were reportedly offered nine-figure pay packages.
[01:00] Nina Park: MuseSpark is the first model in what they call the Avocado series.
[01:05] Nina Park: Thatcher, what do we know about the actual capabilities of this first release?
[01:10] Thatcher Collins: The company is describing it as small and fast by design, with a focus on reasoning through math, science, and health.
[01:17] Thatcher Collins: However, the data from artificial analysis shows a mixed bag.
[01:21] Thatcher Collins: It tied for fourth place on a broad index of tests, showing strength in language and vision,
[01:27] Thatcher Collins: but falling behind OpenAI and Google when it comes to coding and abstract reasoning.
[01:32] Thatcher Collins: It seems like a foundation, but perhaps not the frontier jump some expected for a $14 billion investment.
[01:40] Nina Park: That is a fair assessment, though Mark Zuckerberg did advise investors to expect a rapid trajectory rather than a single definitive leap.
[01:48] Nina Park: One major shift here is the release strategy.
[01:51] Nina Park: Unlike the Lama models, Meta is keeping MuseSpark in a private preview for now, only sharing
[01:57] Nina Park: it with select partners.
[01:59] Nina Park: Thatcher, how does this align with their previous open AI philosophy?
[02:04] Thatcher Collins: It is a clear departure, Nina.
[02:06] Thatcher Collins: By keeping it as a private preview on their own apps first, they are prioritizing their
[02:11] Thatcher Collins: own ecosystem.
[02:13] Thatcher Collins: They are also introducing a feature called Contemplating Mode.
[02:16] Thatcher Collins: It runs multiple agents at the same time to boost reasoning power, similar to what we have seen with Gemini DeepThink.
[02:23] Thatcher Collins: It is designed for multi-step tasks like planning a complex itinerary.
[02:28] Nina Park: And there is a clear push toward monetization here as well.
[02:32] Nina Park: Meta is using MuseSpark to power shopping features within their chatbot,
[02:36] Nina Park: pointing users directly to products.
[02:38] Nina Park: They are betting that applying AI to everyday tasks for their 3.5 billion users
[02:44] Nina Park: will give them a reach that their competitors simply cannot match.
[02:48] Thatcher Collins: Right, but the friction comes in the execution.
[02:51] Thatcher Collins: Alex Wang admitted on social media that there are still rough edges to polish.
[02:55] Thatcher Collins: If they want to compete with GPT-Pro or Gemini, they have to close that gap in coding and complex reasoning.
[03:02] Thatcher Collins: They have promised bigger versions of the avocado series are coming, but for now, this remains a measured first step.
[03:10] Nina Park: It will be a long year of releases for that team.
[03:13] Nina Park: We will continue to track the performance of these new reasoning models as they move into wider release.
[03:19] Thatcher Collins: Thank you for listening to Model Behavior, a neural newscast editorial segment.
[03:25] Thatcher Collins: For more analysis, visit mb.neuralnewscast.com.
[03:31] Thatcher Collins: Neural Newscast is AI-assisted, human-reviewed.
[03:35] Thatcher Collins: View our AI Transparency Policy at neuralnewscast.com.
[03:41] Announcer: This has been Model Behavior on Neural Newscast.
[03:44] Announcer: Examining the systems behind the story.