Meta Plans New Open-Source AI Models Alongside Proprietary Systems
The open-source versions are expected to be derived from more advanced proprietary systems currently in development, internally codenamed “Avocado” and “Mango.”
Meta is preparing to release open-source versions of its next-generation AI models, as the company doubles down on a hybrid strategy that blends openness with proprietary development.
According to a report, the upcoming models—developed under the leadership of Alexandr Wang—will include both closed and open-source variants. The move signals Meta’s continued commitment to making AI tools widely accessible, even as competition intensifies with rivals like OpenAI and Anthropic.
The open-source versions are expected to be derived from more advanced proprietary systems currently in development, internally codenamed “Avocado” and “Mango.”
However, these public releases may not include the full capabilities of their closed counterparts, with potential limitations in features, scale, or post-training refinements.
Meta’s approach reflects a broader balancing act in the AI industry. While open-source models can drive adoption, innovation, and ecosystem growth, companies are increasingly cautious about exposing their most powerful systems due to competitive and safety concerns. As a result, Meta is expected to retain some of its largest and most advanced models as proprietary assets.
The strategy also reinforces Meta’s long-standing push into open AI development through its Llama family of models, which helped popularise open-weight AI systems globally. At the same time, the company is investing heavily in next-generation AI infrastructure and research to close the gap with leading competitors.
