Meta’s AI strategy may be moving into a very different phase. According to Bloomberg’s sources, the company is shifting attention toward a new model codenamed "Avocado," with a release potentially coming next spring.
The reported plan matters because Avocado is expected to be closed, giving Meta a way to sell access directly. That would be a major change for a company known in this space for its open Llama models.
A reported turn away from open Llama models
Meta’s Llama models have been central to its open-model strategy. The source says that approach has reportedly lost momentum inside the company after the disappointing performance of Llama 4.
If Avocado launches as described, Meta would not simply be releasing another model under the same playbook. It would be emphasizing a model built around direct sales, with access controlled by the company.
That distinction is important. An open-model strategy is built around broad availability and outside adoption. A closed model, by contrast, gives the company more control over how access is offered and how the model is commercialized.
What Avocado is expected to be
Avocado is described in the source as a new AI model. Its release is potentially coming next spring, though the report does not present that as a fixed launch date.
The key reported feature is not only the model itself, but the business model around it. Avocado is expected to launch as a closed model, which would let Meta sell access directly.
That would place Avocado in a different category from the open Llama approach. The shift would suggest that Meta sees more value, at least for this effort, in controlling distribution and access than in extending the same open-model strategy.
Why Llama 4 matters to the decision
The report connects the internal change in direction to Llama 4. The source says the open-source approach reportedly lost steam after Llama 4’s disappointing performance.
That does not mean the source says Meta has abandoned every open-model effort. It does mean that Avocado is being framed as a significant strategic pivot, especially because it is expected to be closed and tied to direct sales.
For readers following AI model strategy, the key point is that performance and business model are linked here. If a major open-model release fails to meet internal expectations, leadership may look for a structure that offers more control over the next product.
The role of Alexandr Wang
Management is reportedly placing a large bet on Alexandr Wang, who joined Meta following the company’s deal with Scale AI. The source does not detail his precise responsibilities for Avocado, but it places him inside the broader shift now underway.
That personnel detail matters because the report presents Avocado as more than a routine model update. It is described as part of a bigger change in direction after Llama 4 and amid a move toward a closed, directly sold model.
External models and a strategic tension
The development process reportedly involves several outside models. According to Bloomberg, the team is training Avocado using Google’s Gemma, OpenAI’s gpt-oss, and Alibaba’s Qwen.
That mix is notable for two reasons. First, it suggests Avocado is being developed with help from multiple external model sources. Second, the source points out that using Chinese technology clashes with CEO Mark Zuckerberg’s previous warnings about Chinese censorship.
The report does not say how Meta resolves that tension. It simply places those facts side by side: Avocado is reportedly using Alibaba’s Qwen in training, while Mark Zuckerberg has previously warned about Chinese censorship.
What the reported pivot signals
The clearest takeaway is that Meta may be rebalancing its AI priorities. Open Llama models helped define the company’s public AI posture, but Avocado is described as a closed model built for direct access sales.
That could change how developers, customers, and competitors read Meta’s next move. Instead of treating openness as the defining feature, Meta may be preparing a model where control, access, and monetization are more central.
For now, the facts remain limited to the report: Bloomberg’s sources say Meta is shifting focus to Avocado, the model could arrive next spring, it is expected to be closed, and its training reportedly uses several external models. Within those boundaries, the direction is clear enough: Avocado would mark a serious departure from the open Llama strategy that came before it.