Meta's 'Avocado' AI Model Delayed After Falling Short Of Rivals Like Google Gemini & OpenAI's ChatGPT
· Free Press Journal

Meta's ambitious push to compete at the frontier of artificial intelligence has hit a significant roadblock. The company's next-generation AI model, internally code-named Avocado, has been delayed by at least two months after failing to match the performance of leading models from rivals like Gemini and ChatGPT, according to reporting by The New York Times.
A model that falls short
Visit asg-reflektory.pl for more information.
Internal benchmarks on reasoning, coding, and writing tasks showed that Avocado underperformed against top models from Google, OpenAI, and Anthropic. While the model improved upon Meta's previous generation and outpaced Google's Gemini 2.5, it did not measure up to the more recent Gemini 3.0, released in November. As a result, Meta has pushed Avocado's release from mid-March to at least May.
Google Nano Banana 2 AI Tool Unveiled: Here's What's New & How To Use ItZuckerberg has bet big money on AI
The delay is a notable setback for CEO Mark Zuckerberg, who has wagered billions on Meta becoming an AI powerhouse. The company has committed $600 billion toward data centre infrastructure and projected spending of up to $135 billion in 2026 alone - nearly double what it spent the previous year. In June, Meta also made a $14.3 billion investment in data-labelling firm Scale AI, bringing its founder Alexandr Wang on board as chief AI officer to oversee a new in-house lab called TBD Lab.
With Avocado delayed, Meta's leadership has reportedly discussed temporarily licensing Google's Gemini to power its AI products - though no final decision has been made. The company is also wrestling with whether to release Avocado as an open-source model, a long-standing Meta tradition, or keep it proprietary to better compete commercially. According to the Times, senior executives including Wang leaned toward a closed model over the summer.
'Claude AI Killed My Startup': Founder Explains How One Feature Made Her Business ObsoleteDespite the setback, AI experts noted that model timelines shift frequently and Meta still has room to close the gap. Zuckerberg himself tempered expectations in January, telling investors the first models would demonstrate trajectory more than immediate dominance. A company spokesman echoed this, saying Meta is excited for people to see what it has been building very soon. Meanwhile, a second fruit-coded model, Mango, focused on image and video generation, is also in development, with an even larger follow-up already in the pipeline under the working name Watermelon.