Meta Delays Avocado, Its AI, in Competition with Google and OpenAI

Meta Delays Launch of Avocado

Meta has decided to postpone the launch of its artificial intelligence model, codenamed "Avocado." Initially scheduled for mid-March 2026, the model will not be available until May at the earliest. This decision follows internal testing that revealed disappointing performance compared to industry leaders.

Performance Below Expectations

During internal evaluations, Avocado demonstrated notable shortcomings in logical reasoning, programming, and writing. Although it outperformed Meta's previous model and Google's Gemini 2.5, it failed to compete with the more advanced Gemini 3.0 model. This situation has led Meta's management to temporarily consider acquiring a license to use Google's Gemini model, although no definitive decision has been made.

Future Projects and Investments

Meanwhile, Meta is already preparing a next-generation model named "Watermelon" and developing an image and video generator codenamed "Mango." The company plans to release updates "very soon" and introduce additional models throughout the year.

Meta's Strategy and Investments

Meta initially found success with its Llama models but has lost momentum with Llama 4. To strengthen its position in the AI field, CEO Mark Zuckerberg has made significant investments, including $14.3 billion in Scale AI. Alexandr Wang, CEO of Scale AI, now leads Meta's "TBD Lab," a division dedicated to creating superintelligent AI systems. There are also rumors of a potential shift in Meta's strategy, which may move away from its open-source approach.