Meta released Muse Spark 1.1 on Thursday, claiming the AI model surpasses prior versions of OpenAI, Anthropic, and Google's models on key benchmarks. The update positions Meta closer to the frontier of the AI arms race, though the company acknowledges its model still trails the latest flagship offerings from its competitors. The move underscores Meta's accelerated push under chief AI officer Alexandr Wang, who reorganized the company's AI efforts into Superintelligence Labs.
- What Happened
- The Superintelligence Labs Reorganization
- How Muse Spark 1.1 Stacks Up
- Meta's Broader AI Push
- What This Means for the Industry
What Happened
Meta announced the release of Muse Spark 1.1, the latest iteration of its flagship AI model, touting improvements across coding, video captioning, and reasoning tasks. In a blog post, the company claimed the new model beats Google's Gemini on coding and reasoning benchmarks and outperforms older versions of OpenAI and Anthropic's models on several verticals.
CEO Mark Zuckerberg touted the model's efficiency in a post on X, saying the company's focus is "on delivering strong agentic and multimodal models at very low cost." He added, "More to come soon."
According to Fortune, the tech giant did not claim superiority over the most recent flagship models from OpenAI (GPT-5.6) or Anthropic (Mythos 5 and Fable 5). On at least one coding benchmark, Muse Spark 1.1 still lags behind those frontier systems, per a leaderboard ranking open-source test results.
The release comes amid lingering scrutiny over Meta's past benchmark claims. In April 2025, the company faced accusations of manipulating results on tests for a prior model. A former Meta AI executive denied the allegations at the time, stating the company did not train on test sets.
The Superintelligence Labs Reorganization
Muse Spark 1.1 is the second major model released under Meta Superintelligence Labs, the unit formed after Meta spent $14.3 billion in 2025 to acquire a 49% non-voting stake in Scale AI, the AI data startup founded by Wang. Zuckerberg then tapped Wang as Meta's first chief AI officer, tasking him with overhauling the company's AI strategy.
The reorganization was not smooth. Staffers reported whiplash as Wang and Zuckerberg rebuilt the AI team from the ground up. An Applied AI unit created in March pulled engineers into what some perceived as mind-numbing data collection work. Despite the internal turbulence, Meta's cadence of releases has accelerated since the pivot.
How Muse Spark 1.1 Stacks Up
Meta positions Muse Spark 1.1 as a strong competitor in the mid-tier AI model space, especially for cost-sensitive applications. Zuckerberg stated the model will feature "aggressive pricing" compared to OpenAI and Anthropic's products. The model is now available in public preview via Meta's AI assistant in the Meta AI app and through an API for developers.
However, the company remains coy about direct comparisons to the latest from its rivals. On one open-source coding benchmark, Muse Spark 1.1 ranks below GPT-5.6 and Anthropic's Mythos 5 and Fable 5, indicating Meta has not yet closed the gap at the highest end of performance.
Meta's Broader AI Push
The release of Muse Spark 1.1 coincides with Meta's launch of Muse Image and Muse Video, the first image and video generation models from Superintelligence Labs. These visual models produced impressive results but also sparked controversy when Meta allowed users to apply AI editing effects to public Instagram photos without explicit permission from the original posters, drawing backlash on social media.
Meta's strategy appears to be a dual-track approach: offer capable, low-cost language models for developer and consumer use while building out multimodal generation capabilities. The visual tools could integrate deeply with Meta's social platforms, but the privacy implications may create regulatory headwinds.
What This Means for the Industry
For investors: Meta's heavy spending on AI infrastructure and the Scale AI stake reflects a bet that catching up to frontier labs can be achieved through aggressive investment and a price war. If Meta succeeds in undercutting OpenAI and Anthropic on pricing while delivering adequate performance, it could squeeze margins across the industry. However, the company's willingness to publish benchmark comparisons that acknowledge weaknesses suggests a more transparent posture than in the past.
For competitors: OpenAI and Anthropic now face a well-funded competitor that can afford to lose money on API pricing to gain market share. Meta's integration with social platforms gives it a distribution advantage that pure-play AI labs cannot easily replicate. The pressure will mount on both frontier labs to differentiate through superior performance or specialized vertical offerings.
For the broader tech industry: Lower-cost AI models from Meta could accelerate enterprise adoption, as smaller companies gain access to capable models at reduced prices. The agentic capabilities Meta touts — models that can take actions on behalf of users — signal a shift from chatbots to autonomous digital assistants. This trend could reshape workflows across software, customer service, and data analysis sectors.
Frequently Asked Questions
What is Muse Spark 1.1? Muse Spark 1.1 is the latest version of Meta's large language model, trained for tasks including coding, reasoning, and video captioning. It represents the second major release from Meta Superintelligence Labs.
How does Muse Spark 1.1 compare to OpenAI and Anthropic models? Meta claims it beats older versions of OpenAI and Anthropic's models and Google's Gemini on certain benchmarks. However, it still trails the most recent flagship models — GPT-5.6, Mythos 5, and Fable 5 — on at least one coding metric.
What is Meta Superintelligence Labs? It is a reorganised AI division led by Alexandr Wang, formed after Meta acquired a 49% stake in Scale AI for $14.3 billion. The unit focuses on building frontier AI models and now includes research, product, and infrastructure teams.
Why did Meta acquire a stake in Scale AI? The acquisition gave Meta access to Scale AI's data labeling infrastructure and expertise, as well as the leadership of Wang, who became Meta's first chief AI officer. The deal was a pivot from Meta's earlier open-source AI strategy.
When will Muse Spark 1.1 be fully released? The model is currently in public preview via an API and Meta's AI assistant. Meta has not announced a full general release date but indicated pricing will be aggressive compared to rival offerings.
Is Muse Spark 1.1 open-source? Meta has not specified the licensing for this model. The company originally emphasised open-source AI with its Llama series, but Muse Spark represents a more proprietary approach under the reorganized lab.
Conclusion
Meta's Muse Spark 1.1 shows the company is making real progress in closing the gap with frontier AI labs, though it still trails at the highest end of performance. The combination of aggressive pricing, distribution through Meta's social platforms, and a revamped leadership structure gives the company a credible path to becoming a major AI player. The coming months will reveal whether Zuckerberg's bet on low-cost, capable models can force the industry into a new pricing paradigm.
Reacties