Meta Readies New AI Models to Take On OpenAI and Google

Meta Platforms is developing two next-generation artificial intelligence models, code-named Mango and Avocado, as part of a renewed push to catch up with OpenAI and Google, The Wall Street Journal reported.

The projects, which could debut in the first half of 2026, are being positioned internally as a major inflection point for Meta’s AI ambitions, the Journal report said, citing people familiar with the matter.

Mango is designed as an image- and video-generation model, while Avocado is being built as a large language model focused on advanced text generation and coding. Together, they are expected to anchor Meta’s next wave of consumer and developer-facing next-generation artificial intelligence systems that could challenge the dominance of OpenAI and Google.

The effort is being led by Meta’s Chief AI Officer Alexandr Wang, the founder of data-labeling firm Scale AI, who joined the company earlier this year after Meta acquired a near-majority stake in his startup in a deal valued at more than $14 billion, the Journal reported.

Mango and Avocado will also be the first flagship outputs of Meta Superintelligence Labs, a newly created division designed to accelerate breakthrough AI research and development.

For Meta chief executive Mark Zuckerberg, the initiative reflects a shift from incremental improvements to a broader attempt to reclaim ground lost over the past year, as rivals such as OpenAI and Google set the pace with models including GPT, Sora and the Gemini family.

Mango is expected to target high-quality creative generation, positioning it against tools like OpenAI’s Sora and Google’s Gemini 3 Flash. Avocado, meanwhile, aims to strengthen Meta’s weakest flank: advanced reasoning and coding, areas where its Llama models have lagged behind competitors, the WSJ report said.

Internally, Wang has described Avocado as Meta’s most ambitious large language model yet, one designed not just to generate fluent text, but to solve complex technical problems. 

He has also hinted at the company’s growing focus on so-called “world models,” a shift away from traditional next-word prediction toward systems that attempt to understand how the world works. If successful, this approach could mark a meaningful step toward more perceptive, context-aware AI.

The push follows a sweeping internal reorganization earlier in 2025, when Meta consolidated its AI efforts and placed Wang in charge of a tightly knit group of elite researchers and engineers. 

The company has also aggressively recruited talent, reportedly hiring more than 20 researchers from OpenAI to form what insiders describe as a “super team” of over 50 specialists within MSL.

The timing reflects just how intense the AI arms race has become. Google has rolled out Gemini 3 Flash, positioning it as a faster and more cost-effective model for large-scale use, while OpenAI continues to expand its ecosystem with tools such as ChatGPT Images and its fast-growing video platform, Sora. 

Meta has made smaller moves of its own, including the launch of Vibes, a video-generation tool developed with Midjourney, but those efforts have been quickly overshadowed by rival releases.