Meta's Avocado AI Model Delayed: Falls Short Against Google and OpenAI

Meta's flagship AI stumbles in internal testing

Meta's next-generation AI model, codenamed Avocado, has been pushed back from its expected March 2026 launch to at least May 2026. Internal tests show the model underperforming rivals including Google Gemini, OpenAI GPT-5.4, and Anthropic Claude on logical reasoning, coding, and writing.

A string of setbacks

This isn't the first time Meta has struggled with an ambitious AI launch:

• Llama 4 Behemoth was scrapped after criticism for misleading benchmark results
• TBD Lab – an elite internal unit led by Wang – was created specifically to build Avocado
• TBD Lab's only public release so far has been Vibes, an AI video app

Emergency option: License Google's Gemini?

Most shockingly, Meta is reportedly considering licensing Google's Gemini technology as a stopgap while continuing to develop Avocado. This would mark a dramatic reversal for a company that has invested tens of billions trying to build frontier AI in-house.

15,000 layoffs in the background

The news comes amid Meta's largest-ever layoff round – over 15,000 positions being cut, partly blamed on high AI infrastructure costs.

What this means for enterprise

For CIOs and technology leaders, this is a stark reminder that even the world's most resource-rich companies struggle with frontier AI development. The competition between OpenAI, Google, and Anthropic is so fierce that even Meta – with 1.5 billion daily users as training data – can't keep pace.

The Avocado delay gives Google an opening to cement Gemini's position as the leading open-weight alternative, while Meta regroups.