On Thursday, March 12, 2026, multiple reports from the New York Times and Reuters confirmed that Meta has delayed the launch of its next-generation AI model, codenamed “Avocado,” until at least May 2026.
The model was originally targeted for a mid-March release, but internal testing revealed it is currently “falling between” the performance of Google’s older Gemini 2.5 and the current industry-leading Gemini 3.
The “Avocado” Performance Gap
The delay is a significant strategic hurdle for Mark Zuckerberg’s “Superintelligence” roadmap, for which Meta has committed an infrastructure spend of up to $135 billion this year.
- Benchmark Failure: Internal tests showed that while Avocado is a massive leap over the previous Llama 4 models, it struggled to match rivals like Gemini 3, Claude 4, and GPT-5 in reasoning, coding, and logical “agentic” tasks.
- The “Gemini License” Rumor: In a surprising twist, the NYT reported that Meta’s AI division leaders have discussed licensing Google’s Gemini as a temporary stopgap to power Meta’s AI features while Avocado undergoes more training.
- TBD Lab Origins: The model is the flagship project of TBD Lab, an elite 100-person unit led by Meta’s Chief AI Officer Alexandr Wang (founder of Scale AI).
Strategic Pivot: Focus on “Trajectory”
A Meta spokesperson, Dave Arnold, addressed the reports by shifting the focus to long-term progress:
“Our next model will be good, but more importantly, show the rapid trajectory we’re on, and then we’ll steadily push the frontier over the course of the year as we continue to release new models.”
The Fruit-Themed Roadmap
Avocado is part of a broader series of “fruit-themed” models designed by TBD Lab to tackle different modalities:
| Model Codenamed | Focus Area | Status |
| Avocado | Large Language Model (Reasoning, Coding, Agentic tasks). | Delayed to May 2026. |
| Mango | High-fidelity Image and Video Generation (Sora competitor). | In development; stage unknown. |
| Watermelon | The successor to Avocado (Expected to be the true “superintelligence” contender). | Early planning/pre-training. |
Internal Turmoil & Infrastructure
The delay comes amid reports of internal friction at Meta:
- Leadership Clashes: Reports suggest Alexandr Wang has clashed with Meta veterans like Chris Cox and Andrew Bosworth over how these models should integrate with Meta’s core advertising engine.
- Talent War: Despite hiring over 20 researchers from OpenAI, TBD Lab has seen several high-profile departures recently as equity from the 2024–2025 hiring spree began to vest.
The Market View
Analysts at Investing.com and Seeking Alpha note that while the delay is a PR setback, it signals Meta’s refusal to ship a “sub-frontier” model. With the Stargate supercomputer and Meta’s own MTIA chips scaling up, the company is betting that a May launch with higher performance is better than a mediocre March release.
