Meta was set to launch its new AI model, codenamed Avocado, at the start of this year. The new text-based model is being developed by the company’s Superintelligence Labs (MSL). However, it appears that the tech giant is struggling to meet performance expectations with Avocado and has decided to push back its rollout. This may even open the possibility of Meta licensing Google’s Gemini models temporarily.
According to a report by the New York Times, Meta has pushed back the release to at least May. This decision comes after the system reportedly failed to meet performance benchmarks in internal tests.
Meta’s new AI model falls behind Gemini 3
The report states that in Meta’s internal tests, Avocado failed to outperform Google’s Gemini 3 model that was launched in November last year. However, the model did manage to score higher than Gemini 2.5.
The new model is reportedly behind offerings from companies such as OpenAI and Anthropic too. Though Avocado is also said to outperform Meta’s previous AI models such as Llama 4.
Meta may license Google’s Gemini
As per the report, Meta executives have discussed the possibility of temporarily licensing Google’s Gemini models for its platforms.
The Mark Zuckerberg-led firm provides its AI chatbot, Meta AI, on several of its platforms, including Facebook, Instagram, and WhatsApp. There is also a dedicated Meta AI app.
Licensing Google’s models could allow Meta to retain existing users from going to other LLMs, while the company continues to work on its own models.
This would not be the first instance when a company has licensed a rival’s AI technology. Apple recently announced a billion-dollar deal to use Gemini to power Siri.
Avocado is Meta’s first model from Superintelligence Labs
Meta’s Superintelligence Labs (MSL) is said to be working on its first two AI models, text-based Avocado and image-generation model Mango. The division is led by Alexandr Wang. Meta had acquired Wang’s startup Scale AI for $14.5 billion last year.
Internal reports describe Avocado as an AI model optimised for logical reasoning, software development, and “agentic” behaviour, aiming to function as an autonomous agent capable of multistep planning and execution.
Meta recently acquired Moltbook – the AI-only social media platform that went viral over AI bots demanding a takeover of the world. The company says that it will bring Moltbook’s team to its Superintelligence Labs to bolster agentic AI development,
The company has also committed significant capital expenditures for AI. Meta plans to invest between $115 billion and $135 billion in AI infrastructure this year.
