The New York Times, citing sources, reported that senior members of Meta's newly established super intelligent laboratory are discussing a major strategic shift, abandoning the company's powerful next-generation open-source artificial intelligence model Behemoth in favor of developing a closed model.
It is reported that Meta has completed the training of the Behemoth model, but due to poor internal performance tests, its release was postponed, and the testing work has also been suspended after the new laboratory was launched. If Meta ultimately decides to abandon Behemoth and prioritize the development of a closed model, it will mark a significant shift in its core AI philosophy.
Regarding this, a Meta spokesperson told the media that the company's position on open-source AI "has not changed" and plans to continue releasing leading open-source models, but will mix the training of both open and closed models in the future. The spokesperson did not comment on whether Behemoth might be abandoned. Any strategic change still requires final approval from CEO Mark Zuckerberg.
Zuckerberg once prominently positioned the openness of the Llama series as a core strategy to differentiate from competitors such as OpenAI. However, with Meta investing billions of dollars in AI (including paying large salaries to recruit top talent and building data centers), the company is facing significant profit pressure outside of its advertising business. Despite having top-tier AI research laboratories, Meta lags behind competitors like OpenAI and Google DeepMind in AI commercialization.
Shifting to a closed model would give Meta more control over its technology and a clearer path to profitability. Zuckerberg himself has previously stated that although he supports open source, if a technology is strong enough that open sourcing it would be "unresponsible," he would not open source it.
This potential shift could reshape the global AI landscape. The momentum of open source driven by Meta's Llama series may slow down, and power may shift back to giants with closed ecosystems. At the same time, this will significantly impact the startup ecosystem that relies on open foundation models and may lead to China, which has actively embraced open source AI, taking the lead in the open source field in geopolitical competition.
