A major strategic shift is underway at Meta Platforms, as the company accelerates its formidable artificial intelligence ambitions with the development of two next-generation models codenamed "Mango" and "Avocado," both slated for release in the first half of a highly anticipated year. This dual-track effort is positioned as a high-stakes move to challenge the dominance of rivals like OpenAI and Google, signaling a potential departure from Meta's previously championed open-source philosophy. The projects represent the first major outputs from the newly created Meta Superintelligence Labs (MSL), a specialized division established after a corporate reorganization earlier in the year.[1]

The Mango model is being engineered as a high-stakes, multimodal AI system focused on the generation and understanding of image and video content.[2][3] Mango is designed to enhance Meta's competitiveness in the field of generative visual content, a technology rapidly intensifying as a battleground among the largest tech companies.[4] Meta's Chief AI Officer, Alexandr Wang, has indicated that Mango could integrate directly with the company's powerful social platforms, including Instagram and Facebook, potentially unlocking new avenues for generative advertising, creative tools, and enhanced content creation features like augmented reality filters and tools for Reels and Stories.[2][5] This push follows Meta's earlier foray into the space, such as the AI video tool "Vibes" launched in collaboration with Midjourney, and comes as competitors have seen explosive user growth driven by their own visual AI tools, such as Google's Nano Banana image creation feature.[6][3] The model is specifically aimed at producing high-quality creative content for the vast user and creator base across Meta's ecosystem, directly placing it in competition with cutting-edge models like OpenAI's Sora.[7]

Concurrently, the company is developing Avocado, which represents Meta's most ambitious next-generation large language model (LLM) to date.[7][1] Avocado's core focus is to achieve a generational leap in professional capabilities, particularly in programming and advanced reasoning, aiming for parity with, or even surpassing, frontier LLMs from industry leaders.[2][4][1] Internally, Wang has described Avocado as a model that is intended to "understand the world, not just words," suggesting a leap in foundational comprehension that goes beyond simple text generation.[7] This model is being trained with a significant emphasis on advanced coding, an area that has traditionally been a relative weak point for previous Meta models compared to those developed by key rivals.[1] Meta's leadership views Avocado as its bid for a frontier text-based LLM, critical for powering smarter AI assistants, sophisticated recommendation engines, and more capable in-app "copilots" for tasks ranging from writing code to drafting business workflows.[2][5]

The development of Mango and Avocado marks a pivotal inflection point in Meta’s overall AI strategy, reflecting a decisive move away from the staunchly open-source approach that characterized the highly popular Llama series.[2][4] Multiple reports indicate that Avocado is widely expected to be a closed-source product, a substantial philosophical and business shift for a company that has championed open-source AI as the path forward.[5][8] This strategic pivot is driven by several factors, including intensifying competitive pressure, the need to develop proprietary technology to regain ground against rivals, and a recognition that proprietary models allow the company to monetize its multi-billion dollar investment in compute and talent more directly through gated access, APIs, and paid tiers.[5][9][8] The transition has reportedly caused internal friction, particularly as it follows an underwhelming market reception to Llama 4, prompting an intense re-evaluation of the company's development roadmap under Wang’s leadership.[2][8] The commitment to the proprietary model approach aligns Meta with the business models of OpenAI and Google, who operate closed, revenue-generating systems, fundamentally reshaping the competitive dynamics of the generative AI sector.[10]

Further underscoring the company’s long-term vision, Meta executives have revealed they are in the early stages of researching and developing "world models."[4][1] World models are a cutting-edge area of AI technology designed to learn and internalize the dynamics of the environment, including physics and spatial properties, by processing multimodal data such as visual information.[3][10] This ambitious research, which aims to equip artificial intelligence with an understanding of the real, three-dimensional world, is considered significant for achieving more advanced "embodied intelligence" with applications in robotics, video analytics, and autonomous systems.[3][10][4]

The commitment to this accelerated AI agenda is evident in the scale of Meta's investment and organizational changes. CEO Mark Zuckerberg personally orchestrated a major talent competition to lead the MSL division, poaching over 20 core researchers from OpenAI and assembling a "super team" of seasoned AI specialists.[3][1] The appointment of Alexandr Wang, the 28-year-old founder of Scale AI, to oversee the effort, reportedly cementing his role after Meta invested over $14 billion to acquire a significant stake in his company, signals the aggressive commitment to this post-Llama strategy.[2][1] Meta has also pledged to invest hundreds of billions of dollars into AI infrastructure to support these massive model training efforts.[9][8] The successful delivery of Mango and Avocado in the first half of the year will serve as a crucial test of this costly restructuring, determining whether Meta can successfully pivot from a primarily open-source champion to a leader in developing proprietary, frontier AI that can effectively monetize its vast global user base and challenge the current AI royalty. The ultimate performance of these models will not only dictate Meta’s market position but also influence the broader industry conversation on the future balance between open and closed AI development.[5]