Meta Avocado Delayed: Reading the Strategic Shift

Meta's next frontier model, Avocado, was supposed to launch in March 2026. It's now targeting May or June—and the delay reportedly stems from underwhelming internal benchmarks, particularly against Google's Gemini 2.5 and Gemini 3 in reasoning, coding, and writing.

The more striking detail: Meta is said to be considering temporarily licensing Gemini from Google to power its products. For the company that open-sourced Llama and positioned itself as the democratizer of AI, that would be a remarkable reversal.

The structural pressures are real. Meta's Behemoth model (Llama 4's flagship) has been repeatedly postponed. The company is locked into a $600 billion AI investment commitment through 2028. And DeepSeek's rise is a double-edged story—it proved the power of open weights by leveraging Llama directly, which is simultaneously a validation of Meta's open-source bet and an argument against continuing it.

This is the core tension: open-source builds ecosystems but also arms competitors. Meta gave the world Llama, and the world used it to build things Meta can't yet match. The question now is whether Meta has enough proprietary advantage left to justify the pivot away from openness—or whether it's chasing a closed-source model playbook that Google and OpenAI have already optimized.

From open-source pioneer to reactive follower is a long way to fall. The Avocado delay is a symptom, not the disease.