Zuckerberg signals Meta won’t open source all of its ‘superintelligence’ AI models

Meta CEO Mark Zuckerberg shared her Sight For “Personal Superintegence” on Wednesday, this idea that people will be able to use AI to achieve their personal goals.

A signal in the letter is a signal that Meta is planning to publish AI models by following “Superinteence”.

Zuckerberg writes, “We believe that the benefits of the superpower should be shared as extensively as possible with the world.” “It said that supergence would increase the anxiety of fancy protection. We need to be strict about reducing these risks and be careful about what we choose to open the source.”

The word that is important about the open source. The Histor of the Mater Lama family of open models has been held as the main difference from competitors such as Zuckerberg Open, XAI and Google Dipmind. Meta’s goal was to create Open AI models that were better or better than this closed model. A 2024 lettersZuckerberg writes, “After the next year we hope the future Lama models will be the most advanced in the industry.”

Zuckerberg has left the house to exercise himself on this promise before. “If at one point there is a bit of a qualitative change in what the thing is capable and we think it is not responsible for opening the source, but we will not do it,” he said at one Podcast Last year.

And though many say that the strict definition of the Lama Open Source AI does not fit-partial because Meta has not published its huge training datasets-Zuckerberg’s words point to a possible change in priority: Open Source and Mater Cutting-AJ AI cannot be default.

Meta rivals have a reason to close their models. Buddy models give companies more control over cashing their products. Zuckerberg mentioned last year that Meta business is not dependent on sale in AI models, so “Llama does not stop investing in our earnings, sustainability or research, as it does for closed suppliers.” Meta definitely earns most money from selling internet advertising.

Nevertheless, Open models had the point of view described Before Meta began to feel that it was falling behind the contestants, and the executives became Openai’s GPT -4 model is bound with beating Lama 3 is the time to develop.

Cut in June 2025, when Meta started sincerely its public AGI sprint through investment $ 14.3 billion in scale AI, To achieve the founder of the scale And CEO, and Reorganizing its AI attempt Under a new unit called Meta Superintelligence Lab. Meta has spent Billions of dollars To achieve researchers and engineers from top AI companies and create new data centers.

Recent reports indicate The investment that has managed to break the test on Behemoth, which has the last Lama model, instead focuses on developing a closed model.

With the mission to introduce “Personal Superintendent” to the world of Zuckerberg – he says that a change from rivals is working on “automation all valuable work” – its AI cashing strategy is turning. From Zuckerberg today, it is clear that Meta plans to provide “personal superjection” through its own products with the extended reality glasses and virtual reality headset.

Zuckerberg writes in Wednesday’s letter, “Personal devices like glasses understand our context because they see what we see, we hear what we hear and can contact us all day will become our primary computing device,” Zuckerberg wrote in Wednesday’s letter.

Asked about the closing of its most advanced models possible, the Meta said a spokesperson that the company is committed to opening the source AI and says it is also expected to train the closed sources models in the future.

“Our position in Open Source AI remains unchanged,” a spokesperson said. “We are planning to continue publishing the top open source models. We didn’t reveal everything that we had developed in Histor and we hope to continue training of the open and closed models.”

This article was updated with more information about Mark Zuckerberg’s location on Open AI models.