Meta Unveils Advanced AI Models to Challenge OpenAI and Anthropic

Meta unveiled several new artificial intelligence models on Monday, positioning them on par with the most advanced offerings from competitors such as OpenAI and Anthropic. "We have closed the gap with frontier models, and we are committed to continuing to make significant investments in this field," said Meta founder and CEO Mark Zuckerberg in an interview with the Financial Times. "The new models will enable much more sophisticated behaviors that are at the forefront of what can be done today with AI."

Meta entered the large language model (LLM) space shortly after the AI revolution began roughly four years ago. However, it started from an underdog position due to a lack of significant prior experience, unlike Google, which maintained extensive research and development activities. Consequently, Meta's initial models lagged significantly behind industry leaders, relying primarily on their open-source nature for an advantage. About a year and a half ago, Zuckerberg initiated a strategic overhaul, beginning with a 49% acquisition of ScaleAI for $14 billion and the recruitment of its founder, Alexandr Wang, to lead Meta's AI division. Simultaneously, the company launched a massive talent acquisition campaign, offering compensation packages reportedly reaching hundreds of millions of dollars.

New Models and Commercial Shift

These efforts are now bearing fruit with the introduction of two new models: Muse Wonder and Muse Wonder Full. According to Meta, the models achieved scores comparable to the most advanced systems from Anthropic and OpenAI in benchmarks measuring reasoning, writing, and coding capabilities, and even outperformed them in specific tasks. The models are available via Meta's API and are integrated into its coding platform, Muse Code, with plans to power the company's popular AI assistant, Muse, in the near future. Unlike the company's earlier releases, these are closed models, with their most advanced versions sold directly to businesses. This move places Meta in direct competition with Anthropic and OpenAI, which derive their revenue from enterprise sales.

"Reaching this point required a massive amount of computing power and investment, along with a great deal of hard scientific work from the team that worked on this for a long time," Zuckerberg told the Financial Times. "We are already working on the next phase, and model development will likely feel much faster from now on."

Wang told Bloomberg: "These are true frontier models. We genuinely think they compete with the best models out there, especially when it comes to agentic work for personal or professional use. For anyone who is skeptical, I say 'try the models and judge for yourself.'"

Vision of Personal Superintelligence

Zuckerberg also discussed Meta's approach to the industry, stating that the company is focused on a vision of "personal superintelligence" for consumers, contrasting with Anthropic and OpenAI's pursuit of artificial general intelligence capable of replacing humans in most tasks. "Everyone else is focused on knowledge work and automation; we are much more focused on personal empowerment. We want to build things that do good in the world," he said. Despite recent incidents where advanced AI models bypassed online security to perform unrequested tasks, Zuckerberg dismissed catastrophic risks, stating he believes responsible engineering and safety mechanisms are sufficient.