22 hours ago
2
Meta is throwing serious computational firepower at its next frontier AI model. Watermelon, the company’s upcoming proprietary system, reportedly uses an order of magnitude more compute than its predecessor Avocado, according to Meta AI chief Alexandr Wang.
Meta had been developing Avocado, a text-focused AI model, with plans to ship it by the end of 2025. Those plans didn’t survive contact with reality. Avocado fell behind Google’s Gemini 3.0, which launched in November 2025, in critical areas like reasoning and coding. The result: Meta pushed Avocado’s release to at least May 2026.
Wang has reportedly claimed that Watermelon has caught up to OpenAI’s improved GPT-5.5. The ten-times compute increase tells you how Meta plans to get there. Meta’s model pipeline doesn’t stop there either. The company also has a model codenamed Mango in development, designed specifically for image and video generation.
In June 2025, Meta invested $14.3 billion in Scale AI, the data infrastructure company founded by Wang. That investment came alongside Wang’s appointment as Meta’s chief AI officer. The shift marks a notable departure from Meta’s previously open-source Llama model series toward closed, proprietary systems like Avocado and Watermelon.
The AI-crypto intersection has become one of the most consequential themes in digital asset markets. AI tokens, decentralized compute networks, and GPU marketplace protocols all trade on the fundamental thesis that AI compute demand will keep growing exponentially. Decentralized compute platforms like Render, Akash, and io.net have positioned themselves as alternatives to centralized cloud providers for AI workloads.
Scale AI’s core business involves preparing training data for AI models. A $14.3 billion investment in Scale AI validates the market opportunity for decentralized data labeling and data marketplace protocols in crypto, even if it simultaneously strengthens a centralized competitor.
For investors, the key risk is timeline uncertainty. Meta’s Avocado delays illustrate how unpredictable frontier AI development remains. The company expected to ship by late 2025 and missed. Watermelon has no publicly disclosed release date or detailed parameter specifications. If Watermelon truly matches GPT-5.5 performance, it could reshape the AI landscape in ways that affect which infrastructure providers capture the most value. But if it underwhelms, as Avocado did against Gemini 3.0, questions about the efficiency of Meta’s compute-heavy approach could dampen enthusiasm for the entire “more compute equals better AI” thesis that underpins much of the AI token narrative.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) · 