Meta delays Mango and Avocado AI models to the first half of 2026

In short: Meta told employees it will release its next two major AI models, Mango and Avocado, in the first half of 2026, later than planned.

What happened

Meta Platforms has delayed two upcoming AI models it has been working on internally. The models are called Mango and Avocado. Mango is meant for images and video, and Avocado is meant for advanced text and coding.

Meta had previously expected Avocado to be ready by the end of 2025. The company now says both models are targeted for the first half of 2026. The update was shared internally by Meta’s Chief AI Officer, Alexandr Wang, during a question and answer session with employees.

The delay comes after Meta’s earlier model, Llama 4, had a rocky rollout in April 2025. Reports said it did not score as well as expected on common tests and struggled in coding tasks. Meta also faced criticism from developers after it submitted an improved version to public scoreboards instead of the version most people could actually use.

Meta has spent heavily on AI. That includes a $14.3 billion investment for a near-majority stake in Scale AI and hiring more than 20 researchers from OpenAI, according to the report. CEO Mark Zuckerberg has also been rebuilding Meta’s AI teams, while the company expects overall spending to rise.

Why it matters

For most people, this means Meta is still not ready to offer a widely used AI assistant that competes directly with tools like ChatGPT or Google’s Gemini (a chatbot that answers questions). Even with large spending, many of Meta’s AI efforts are still mostly research and demos, not everyday features inside Instagram, WhatsApp, or Facebook. The next year will show whether Meta can turn that spending into tools people actually use.

Source: NYTimes