Avocado on Ice: Can Meta Afford to Pause While Google and OpenAI Sprint Ahead?

"Previous Llama models were open-source, freely available for developers worldwide to download, fine-tune, and deploy. That approach fueled rapid community adoption and positioned Meta as a democratizing force in AI. Avocado, however, is being developed as a proprietary, closed-source system - much like those from OpenAI, Google, and Anthropic."

"Proprietary models allow higher profit margins because Meta can control distribution, charge for premium access, or integrate the technology exclusively into its own products across Facebook, Instagram, and WhatsApp. A wider economic moat also emerges: rivals cannot simply fork the code and leapfrog Meta's progress."

"The New York Times reported that Meta has pushed the rollout back to at least May - and possibly June - after internal tests showed Avocado lagging behind rivals such as Google's Gemini 3.0 in critical areas like reasoning, coding, and writing."

Meta is developing Avocado, a next-generation AI model intended to succeed its Llama series. Originally targeted for Q1 launch, the rollout has been postponed to May or June after internal testing showed Avocado underperforming compared to competitors like Google's Gemini 3.0 in key areas including reasoning, coding, and writing capabilities. Despite outperforming Meta's previous models, the performance gap with frontier competitors was deemed unacceptable. This delay has raised questions about Meta's substantial AI infrastructure investments. Notably, Avocado represents a strategic shift from Meta's traditional open-source approach to a proprietary, closed-source model similar to offerings from OpenAI, Google, and Anthropic, enabling higher profit margins and stronger competitive advantages.

Read at 24/7 Wall St.