TL;DR:

According to Alexander Wang, Meta’s AI Chief: Meta is reportedly developing Mango, a new AI model focused on image and video generation, and Avocado, a text model described as being stronger at coding. Both are expected to target a first half of 2026 release window. The same internal discussion also referenced early work on “world models,” a clue that Meta is thinking beyond chat and into AI that understands environments.

Avocado: text, with a not so subtle focus on coding

Avocado is described as a text-based model, with reporting highlighting improved coding ability.

That matters because coding ability is not just “developer stuff.” In 2026, coding strength is how you build:

• agents that can take multi step actions,
• internal automation that replaces repetitive ops work,
• product features that ship faster than your competitors can copy.

Coding is leverage. Meta knows it.

Why these “world-class models” matter for Meta?

In mid 2025, Reuters reported Meta reorganized its AI work under a new division called Superintelligence Labs, led by Alexandr Wang, with Nat Friedman co leading and overseeing AI products and applied research.

That reorganization matters because it changes incentives. When a company creates a dedicated “war room” org, it usually means two things:

1. the old structure was too slow,
2. leadership wants fewer excuses and faster shipping.

In other words, Mango and Avocado are not side bets. They look like outputs from a new operating system inside the company.

Meta already started training users for AI video: Vibes

On September 25, 2025, Meta announced Vibes, an AI video feed inside the Meta AI app and on meta.ai, where people can create, remix, and share short AI generated videos.

That product is worth mentioning for one simple reason: it shows Meta is actively trying to make AI video feel normal.

It also quietly answers a question you might have had. If Meta is building Mango, where does it get used? Vibes is an obvious place, because Vibes is already built around the habit of “see something, remix it, post it.”

What this means for marketers, creators, and teams running Meta Ads

You do not need to predict model architecture to make this useful. You just need to watch the right outcomes.

Here are four practical things to track over the next 6 to 12 months:

• Whether Avocado becomes an API product. If Meta prices it, that tells you it sees real enterprise demand, not just consumer chat usage.
• Where Mango shows up first. If it lands inside creator tools, it’s a culture play. If it lands inside ads tooling, it’s a revenue play.
• How controllable the output is. The market is tired of “cool demos.” Winning tools let you keep characters consistent, make local edits, and iterate quickly. (This is why Google’s “nano banana” style tools got attention.)
• What Meta does about quality spam. AI video feeds can become “AI slop” fast, and distribution platforms eventually have to choose between growth and trust.

Bottomline

Mango and Avocado are still just reported codenames. The real test is shipping, and shipping at quality.

But Meta’s direction is already visible. It is moving from “AI that answers” to AI that creates, and possibly AI that sells. If Mango powers a serious visual engine in 2026, the creative workflow for Meta ads and content will shift again. If Avocado goes closed and paid, Meta’s open model narrative changes overnight.