Meta is postponing the launch of its new AI model Avocado until at least May. The model performs better internally than its predecessor, but does not meet the bar set by Google Gemini 3.0 from November 2025. In the meantime, Meta’s AI division is considering temporarily licensing Gemini to power AI products.
This is according to sources speaking to the New York Times. Avocado was initially planned for release this month. Avocado is being developed by TBD Lab, an AI lab within Meta, founded by Alexandr Wang, former CEO of Scale AI.
In internal tests across reasoning, coding, and writing, Avocado lags behind the top models from Google, OpenAI, and Anthropic. The model performs better than Meta’s previous model, Llama 4, which failed to meet expectations last year. It does perform better than Google’s Gemini 2.5 from March 2025. However, Gemini 3.0 from November is still a step too far.
Meta is investing heavily in its AI ambitions. The company has budgeted up to $135 billion this year, nearly double last year’s amount. Yet money alone is not enough. Avocado’s pre-training was completed at the end of last year, but the post-training phase proved more complex than expected.
Gemini license and internal tensions
Because Avocado is not yet ready, managers at Meta’s AI division are discussing the possibility of temporarily licensing Google’s Gemini for AI products on Facebook, Instagram, and WhatsApp. A final decision has not yet been made.
Internally, there is also a discussion about whether Avocado will be released as open source. Meta has always supported open source models, but Zuckerberg and Wang considered keeping the new model closed last summer.
In doing so, Wang clashed with CPO Cox and CTO Bosworth over the role of AI in the advertising business. Last week, Meta announced it would create a separate AI engineering team under Bosworth that would collaborate with Wang’s division.
Tip: Meta CTO: Superintelligence Labs delivers first high-quality models