Meta reportedly delays the launch of its new AI model because it’s just not that good
Meta’s AI leaders have “instead discussed temporarily licensing Gemini to power the company’s AI products, though no decisions have been reached,” according to The New York Times.
Here’s The New York Times with something that Meta CEO Mark Zuckerberg probably wishes had never seen the light of day:
Per the NYT, the social media giant is postponing the release of its new foundational AI model, originally planned for this month, until at least May, citing three people with knowledge of the matter.
The model, which is code-named “Avocado,” reportedly did not perform as well as offerings from Google, OpenAI, and Anthropic “on internal tests for reasoning, coding, and writing.”
In what’s seemingly a concession to Google’s prowess, Meta’s AI leaders “had instead discussed temporarily licensing Gemini to power the company’s AI products, though no decisions have been reached,” according to the report.
Gemini 3.0’s launch was extremely warmly received by the public and the stock market, resulting in a halo effect that saw companies tied to its supply chain soar while firms with lots of exposure to OpenAI sank.
Meta’s prior model, Llama 4, was also plagued by delays and performance issues. Soon thereafter, the firm began bolstering its bench with a high-profile hiring spree, including onboarding Scale AI founder Alexandr Wang after investing $14.3 billion into the startup. Earlier this year, Meta CTO Andrew Bosworth told the press that these new models under development were “very good.”
The social media giant’s capital expenditure over the past two years has totaled nearly $107 billion, as it and other so-called hyperscalers and foundational model companies race to build better AI models and monetize their new capabilities.
But based on this report, aggressively accumulating talent and deploying compute does not ensure that your models will be best-in-class.
