Media reports indicate that Meta is developing a next-generation image and video AI model codenamed Mango, alongside a large language model codenamed Avocado, which focuses on enhancing programming capabilities. Last week, it was reported that Meta is working on Avocado, potentially adopting a proprietary rather than open-source model, marking a shift from open-source to closed-source approaches.
Social media giant $Meta Platforms (META.US)$ has recently made headlines for intensifying its efforts in the AI competition, reflecting a strategic shift from an open-source model to pursuing cutting-edge models. This marks a significant transition for the company.
On Thursday, October 18, Eastern Time, media reports revealed that Meta’s Chief AI Officer, Alexandr Wang, disclosed during an internal Q&A session last Thursday that Meta is developing a next-generation image and video AI model codenamed Mango, as well as a next-generation large language model (LLM) codenamed Avocado, which are expected to be released in the first half of 2026.
Wang stated that one of the key focuses of the Avocado model is enhancing programming capabilities, while the company is also in the early stages of researching and developing world models. A world model refers to an AI technology that learns about environments by absorbing visual information.
This news further confirms Meta’s strategic adjustment in the AI field. Last week, it was reported that Meta is developing a new frontier AI model named Avocado, which may adopt a proprietary rather than open-source model. This contrasts sharply with Meta’s previously promoted open-source Llama series.
To drive AI research and development, Meta restructured its AI team this summer, appointing Alexandr Wang to lead the newly established Superintelligence Labs. CEO Zuckerberg also personally recruited over 20 researchers from OpenAI to form an AI expert team of more than 50 members.
Dual Model Strategy: Advancing Image Generation and Language Capabilities
According to reports on Thursday this week, Meta is simultaneously advancing the development of two core AI models. The image and video model Mango aims to strengthen Meta’s competitiveness in generative AI, while the text model Avocado focuses on enhancing key capabilities such as programming.
Image generation has become a critical battleground for competition among major AI companies.
At the end of August this year, Google launched Nano Banana, an AI image generation and editing tool based on the Gemini 2.5 Flash model, driving Gemini’s monthly active users from 450 million in July to over 650 million by the end of October.
On September 25, Meta launched Vibes, an AI video generator developed in collaboration with Midjourney. Within less than a week, OpenAI released its own video generation application, Sora.
OpenAI CEO Sam Altman emphasized the importance of AI image generation for consumers during a meeting with reporters last week, calling it a key interest for many users and a “sticky” feature that keeps them coming back.
Strategic Shift: From Open Source to Proprietary Model
According to last week’s reports, Meta’s AI strategy is undergoing a significant transformation. Many within the company had originally anticipated the release of the Avocado model by the end of this year; however, the plan has been postponed to the first quarter of 2026. Sources revealed that the model may adopt a proprietary model, meaning external developers will not have free access to download its weights and related software components.
One catalyst for this shift, according to the report, was the lukewarm reception of Llama 4 following its release in April. Additionally, the R1 model launched by Chinese AI newcomer DeepSeek incorporated elements of the Llama architecture, which left some Meta employees dissatisfied and further highlighted the risks associated with an open-source strategy.
Zuckerberg predicted last year that the Llama series would become the “most advanced” model in the industry and specifically discussed Llama during the earnings call in January. However, in the latest earnings meeting in October, he mentioned the brand only once.
In June this year, Meta made a substantial investment of $14.3 billion in Scale AI, bringing in the founder of the unicorn, Alexandr Wang, and announced during the third-quarter earnings report at the end of October that it would raise its capital expenditure guidance for this yearto $70 billion to $72 billion.to USD 70 billion to 72 billion.
World Models: A New Frontier for AI Understanding of the Physical World
As previously introduced by The Paper, world models draw inspiration from the human mind’s conceptual framework. Abstract information acquired through sensory input is transformed into concrete understanding of the surrounding world in the brain. Based on these models, the brain makes predictions about the world, influencing perception and action.
NVIDIA points out that world models are neural networks used to understand the dynamics of the real world, including physical and spatial properties. They can use inputs such as text, images, video, and motion to generate simulated videos of actual physical environments, providing artificial intelligence with the ability to comprehend the real three-dimensional physical world. This holds significant implications for the realization of embodied intelligence.
However, world models face significant technical challenges. Compared to the computational resources currently used for generative models, training and running world models require immense computing power. World models also suffer from hallucination issues and can internalize biases present in the training data. If these obstacles are overcome, world models could bring breakthroughs to robotics and AI decision-making, enabling AI to develop an understanding of its environment and reason out potential solutions.
Looking to pick stocks or analyze them? Want to know the opportunities and risks in your portfolio? For all investment-related questions,just ask Futubull AI!
Editor/Stephen
