Meta Shifts from Open AI Stance, Signaling Strategic Pivot

Senior leaders at Meta's new Superintelligence Lab have reportedly discussed moving away from the company's powerful open-source AI model, Behemoth, in favor of developing a closed model, according to The New York Times.

Sources informed The Times that while Meta completed training for Behemoth, its release was postponed due to lackluster internal performance. Testing on the model reportedly stopped when the Superintelligence Lab was established.

These conversations are preliminary. Any final decision would require approval from Meta CEO Mark Zuckerberg, and a company spokesperson informed TechCrunch that Meta's stance on open-source AI remains "unchanged."

"We plan to continue releasing leading open source models," the spokesperson stated. "We haven't released all of our past developments and anticipate training a combination of open and closed models moving forward."

The spokesperson offered no remarks on Meta's potential shift from Behemoth. Should Meta prioritize closed-source models, it would indicate a substantial philosophical shift for the organization.

Although Meta uses advanced closed-source models internally, such as those for its Meta AI assistant, Zuckerberg had championed open source as a key component of its external AI strategy—a method to accelerate AI advancement. He actively contrasted the openness of the Llama models with competitors like OpenAI, which he publicly criticized for becoming less open after its Microsoft partnership. However, Meta faces pressure to generate revenue beyond advertising as it invests billions in AI.

These expenses include substantial signing bonuses and salaries reaching nine figures to attract top researchers, constructing new data centers, and funding the immense costs of developing artificial general intelligence (AGI), or "superintelligence."

Save up to $475 on your TechCrunch All Stage pass

Build smarter. Scale faster. Connect deeper. Join leaders from Precursor Ventures, NEA, Index Ventures, Underscore VC, and more for a day filled with strategic insights, workshops, and valuable networking.

Save $450 on your TechCrunch All Stage pass

Build smarter. Scale faster. Connect deeper. Join leaders from Precursor Ventures, NEA, Index Ventures, Underscore VC, and more for a day filled with strategic insights, workshops, and valuable networking.

Despite operating one of the world's premier AI research labs, Meta still falls behind competitors like OpenAI, Anthropic, Google DeepMind, and xAI in commercializing its AI developments.

Should Meta prioritize closed models, it might signal that its open-source advocacy was a strategic maneuver rather than an ideological commitment. Zuckerberg's previous remarks suggest some uncertainty about fully committing to open-sourcing all Meta models. On a podcast last summer, he commented:

"We are clearly very supportive of open source, but I haven't pledged to release everything we create. I'm generally inclined to believe open sourcing benefits the community and us, as we gain from the resulting innovations. However, if the model's capabilities undergo a fundamental change and we deem open-sourcing irresponsible, we will not proceed. Predicting these scenarios is challenging."

Closed models would provide Meta with greater control and additional monetization avenues—particularly if the company believes its acquired talent can deliver competitive, top-tier performance.

This potential shift could also transform the AI landscape. The open-source momentum, largely fueled by Meta and models like Llama, could diminish, even as OpenAI prepares to release its long-awaited open model. Influence could return to major players with closed ecosystems, while open-source development may persist largely through community-led initiatives. The repercussions would extend across the startup ecosystem, especially affecting smaller companies specializing in fine-tuning, safety, and model alignment that depend on access to open foundation models.

Globally, Meta stepping back from open source might create opportunities for China, which has adopted open-source AI—including models like DeepSeek and Moonshot AI—to build domestic capabilities and increase its international influence.