Meta Spent Billions on AI. Its Next Big Models Still Aren’t Ready
The Chief AI Officer told employees that its two most important AI models won’t ship until next year.
Mark Zuckerberg has spent the past year tearing down and rebuilding Meta’s entire AI operation. He personally poached more than 20 researchers from OpenAI. He dropped $14.3 billion to secure a near-majority stake in Scale AI and brought in its 28-year-old founder, Alexandr Wang, as Chief AI Officer.
And yet, on Thursday, during an internal Q&A with employees, Wang had to deliver a blunt update: Meta’s two most important AI models won’t ship until next year.
Users, developers, and creators waiting to see what Meta’s AI push would actually look like in products they use would wait a while longer.
What Meta Was Supposed to Ship and When
The two models sit at the heart of Meta’s AI ambitions:
- “Mango” focused on images and video
- “Avocado" is designed for advanced text and coding
Both were outlined by Wang alongside Chief Product Officer Chris Cox, with a revised target of the first half of 2026. Internally, people at Meta were expecting Avocado before the end of this year. It has now slipped into early 2026.
For anyone hoping Meta would quickly catch up to tools like ChatGPT or Google’s Gemini, this is another reminder that the gap isn’t closing anytime soon.
Why This Delay Matters to Users, Not Just Meta
This isn’t just an engineering hiccup. It’s a relevance problem.
While ChatGPT processes billions of prompts daily and Google keeps embedding Gemini deeper into Search, Docs, and Android, Meta’s AI still largely lives behind the scenes, in research posts, demos, and vague promises.
For everyday users, that means:
- No clear breakthrough AI features in Instagram, WhatsApp, or Facebook yet
- No compelling Meta-built alternative to ChatGPT for writing, coding, or productivity
- No clarity on when Meta’s AI will meaningfully change how its apps work
The longer Mango and Avocado are delayed, the harder it becomes for Meta to convince users that its AI investments will translate into tools they actually care about.
Oluwajeminipe Fasheun-Motesho
Meta May Be Abandoning OpenAI
This is the part that changes everything.
According to multiple reports, Avocado may launch as a closed model, meaning Meta would tightly control access and sell usage rather than release it freely. If that happens, it would mark Meta’s sharpest break yet from the open-source Llama strategy it has championed for years.
Meta spent years positioning itself as the company pushing “open AI.” It celebrated hundreds of millions of Llama downloads and framed openness as its moral and strategic edge over rivals.
Now, that stance appears to be quietly eroding.
For developers who adopted Llama because it was open, this raises uncomfortable questions about whether Meta can still be trusted as a long-term platform partner.
What Went Wrong With Llama 4
The pivot didn’t come out of nowhere.
Llama 4’s April rollout was widely seen as a failure. The model underperformed on key benchmarks, lagged behind competitors on coding tasks, and triggered developer backlash. Meta later admitted it had submitted a specially optimised version to public leaderboards rather than the model users actually received.
Internally, that episode reportedly triggered a major shake-up — corporate language for panic.
From a user perspective, it exposed a deeper issue: Meta’s AI narrative hasn’t matched the reality of what people can actually use.
Big Ideas, Unclear Payoff
Wang told employees that Avocado will focus heavily on coding, historically a weak area for Meta’s models. He also pointed to work on so-called “world models,” AI systems trained on visual data rather than just text.
It all sounds ambitious. It also sounds like a company still searching for a clear identity in a race where competitors already have one.
For consumers, ambition only matters if it shows up in usable products. Right now, Meta’s AI story is still mostly internal.
Billions Spent, Questions Still Open
The timing adds pressure. Meta has raised its 2025 capital spending to $70–72 billion, with CFO Susan Li warning that spending will rise again in 2026.
Wall Street wants returns. Employees want certainty. Developers want consistency. Users want tools that work.
What they’re getting instead are fruit-themed codenames, sliding timelines, and a possible closed-source pivot that undercuts Meta’s biggest AI talking point.
The Question That Actually Matters
The real question isn’t whether Mango and Avocado will eventually ship.
It’s whether they’ll still matter by the time they do — and whether Meta can turn years of spending and reshuffling into AI products people actually choose to use.
For now, Meta’s AI future remains expensive, delayed, and unresolved.
Ogbonda Chivumnovu
