Meta delays its Avocado AI model launch from March to at least May after internal tests show it underperforms against Google's Gemini 3.0 and rivals from OpenAI and Anthropic. The model outperforms Meta's previous AI and Google's older Gemini 2.5 from March, but falls short of current industry leaders. Meta's AI division is now considering temporarily licensing Google's Gemini to power its products while continuing development.
Why it matters
Meta's $115-$135 billion AI infrastructure investment faces a credibility test as the company struggles to match competitors in foundational models that power chatbots, coding tools, and content generators. Falling behind in AI capabilities threatens Meta's ability to recruit top AI talent and maintain competitive products. The potential Gemini licensing deal signals that even tech giants with massive resources may need to rely on third-party AI rather than building everything in-house.
What to do
Reassess your AI vendor strategy to include multiple providers rather than betting on a single platform, as even Meta considers licensing competitors' models. Prioritize AI solutions based on current performance benchmarks rather than vendor promises, and build flexibility into contracts to switch providers as capabilities evolve.