Meta is quietly preparing a major reset of its artificial intelligence strategy. The company is developing a new generation of image, video, and text AI models and is targeting a release window in the first half of 2026.
The initiative is being led by Meta’s Superintelligence Lab, now its most critical AI unit.
The renewed push follows a turbulent year marked by restructurings and leadership changes. Internally, Meta sees this effort as a chance to redefine its long-term AI direction. The new models aim to compete directly with rivals such as Google and its emerging Nano Banana models.
Mango and Avocado Form the Core of Meta’s New AI Roadmap
Meta is developing an image and video model codenamed Mango. Alongside it, engineers are building a text-based model known internally as Avocado. Senior leaders recently presented both projects during a company-wide internal session.
Meta expects these models to do more than close competitive gaps. Instead, leadership wants them to reshape how AI powers products and research. The roadmap positions these systems as foundational technologies for future platforms.
The Avocado text model focuses heavily on coding-related tasks. Meta has lagged rivals in developer-focused AI performance. Improving coding capabilities is now seen as essential for enterprise credibility.
Meta Bets on World Models and Contextual Intelligence
Beyond text and media generation, Meta is investing in so-called world models. These systems aim to understand visual environments more holistically. Rather than memorizing scenarios, they focus on reasoning and planning.
Meta believes future AI must move beyond pattern prediction. The company wants systems capable of contextual understanding and autonomous action. This philosophy reflects a broader shift inside Meta’s AI research teams.
Internal Pressure Mounts as Meta Trails Rivals
Despite massive scale, Meta still lacks a standalone AI product users actively choose. Its AI assistant reaches billions mainly through default placement in social apps. However, scale has not translated into trust or clear user preference.
Internally, Meta’s AI division has faced repeated reorganizations. Leadership roles have shifted, and teams have been restructured multiple times. Although Meta recruited aggressively from rival labs, retention has proven uneven.
Several high-profile Superintelligence Lab hires have already departed. These exits have raised concerns about execution and continuity. As a result, Mango and Avocado now carry outsized strategic importance.
Meta does not view these models as incremental upgrades. Instead, they are meant to anchor the company’s next AI chapter. Their success could shape everything from developer tools to future consumer products.