[Rumor] Meta Secret Model Avocaco Slips to May Following Underwhelming Benchmarks.

 

Meta Secret AI "Avocaco" Faces Delays: Internal Tests Struggle to Outperform Google’s Gemini 3.0

Reports late last year suggested that Meta was developing a formidable new AI model codenamed "Avocaco." Intended to succeed the Llama series, Avocaco was rumored to be a proprietary, closed-source model designed to bridge the gap with industry rivals. While an initial launch was expected within Q1 2026, new information suggests the project has hit a significant roadblock.

Performance Gap and Internal Friction

According to The New York Times, citing three internal sources at Meta, the launch of Avocaco has been pushed back to at least May 2026. Despite being Meta’s most powerful model to date, benchmark results have reportedly been underwhelming when compared to current market leaders.

Sources indicate that while Avocaco successfully outperforms Google’s older Gemini 2.5, it has failed to surpass the capabilities of Gemini 3.0 (released late last year). This performance lag has sparked a heated debate within Meta AI. Some executives have reportedly proposed a radical short-term solution: licensing Gemini from Google to ensure other Meta departments can continue developing AI-driven services without further delays.

Meta’s Official Stance

A spokesperson for Meta addressed these reports, stating that the company remains on track with its roadmap and expects to release its next-generation AI models within this year.

Meta's potential shift from an open-source approach (like Llama) to a closed model like Avocaco reflects pressure from investors. Developing high-level AI involves enormous costs, so Meta may need to create a "tariff barrier" or retain its best features to gain a true competitive advantage.

The difficulty in overcoming Gemini 3.0's hurdle stems from Google's use of more sophisticated native multimodal video and audio processing technology. If Avocaco continues to struggle with reasoning or data accuracy, this will be a significant obstacle.

The proposal to lease Gemini is one of the most shocking developments in Silicon Valley. If it materializes, it would transform the relationship between competitors into "coopetition." Conversely, it could represent a technological surrender that Mark Zuckerberg likely wants to avoid.

Rumors suggest that part of the delay is due to fine-tuning the software to work with Inference.net and the new chips. If Avocaco is delayed too long, Meta could be at a disadvantage in the fierce competition for "AI agents." 

 

 

Adobe Posts Record $6.4B Revenue as CEO Shantanu Narayen Prepares for Succession. 

 

Source: The New York Times

💬 AI Content Assistant

Ask me anything about this article. No data is stored for your question.

 

Meta Secret AI "Avocaco" Faces Delays: Internal Tests Struggle to Outperform Google’s Gemini 3.0

Reports late last year suggested that Meta was developing a formidable new AI model codenamed "Avocaco." Intended to succeed the Llama series, Avocaco was rumored to be a proprietary, closed-source model designed to bridge the gap with industry rivals. While an initial launch was expected within Q1 2026, new information suggests the project has hit a significant roadblock.

Performance Gap and Internal Friction

According to The New York Times, citing three internal sources at Meta, the launch of Avocaco has been pushed back to at least May 2026. Despite being Meta’s most powerful model to date, benchmark results have reportedly been underwhelming when compared to current market leaders.

Sources indicate that while Avocaco successfully outperforms Google’s older Gemini 2.5, it has failed to surpass the capabilities of Gemini 3.0 (released late last year). This performance lag has sparked a heated debate within Meta AI. Some executives have reportedly proposed a radical short-term solution: licensing Gemini from Google to ensure other Meta departments can continue developing AI-driven services without further delays.

Meta’s Official Stance

A spokesperson for Meta addressed these reports, stating that the company remains on track with its roadmap and expects to release its next-generation AI models within this year.

Meta's potential shift from an open-source approach (like Llama) to a closed model like Avocaco reflects pressure from investors. Developing high-level AI involves enormous costs, so Meta may need to create a "tariff barrier" or retain its best features to gain a true competitive advantage.

The difficulty in overcoming Gemini 3.0's hurdle stems from Google's use of more sophisticated native multimodal video and audio processing technology. If Avocaco continues to struggle with reasoning or data accuracy, this will be a significant obstacle.

The proposal to lease Gemini is one of the most shocking developments in Silicon Valley. If it materializes, it would transform the relationship between competitors into "coopetition." Conversely, it could represent a technological surrender that Mark Zuckerberg likely wants to avoid.

Rumors suggest that part of the delay is due to fine-tuning the software to work with Inference.net and the new chips. If Avocaco is delayed too long, Meta could be at a disadvantage in the fierce competition for "AI agents." 

 

 

Adobe Posts Record $6.4B Revenue as CEO Shantanu Narayen Prepares for Succession. 

 

Source: The New York Times