Meta's New AI Model Watermelon Surpasses GPT-5.5
Business Insider reports that Alexandr Wang, CEO of Scale AI, revealed in an internal meeting that Meta's upcoming AI model, codenamed Watermelon, has reached the level of OpenAI's GPT-5.5.
Watermelon uses an order of magnitude more computational resources than the previous Avocado model, highlighting Meta's aggressive strategy in AI training investment.
This statement reflects that computational scale has become a key variable in the competition for large models, with Meta accelerating its pursuit of cutting-edge capabilities through massive resources.
Source: Public information
ABAB AI Insight
Alexandr Wang, as the founder of Scale AI and a long-time contributor to top labs like OpenAI, indicates that Meta is significantly increasing its computational investment through the Watermelon project, suggesting that the previous Avocado model may have been a transitional model.
In terms of capital strategy, Meta is leveraging advertising cash flow to support substantial GPU purchases and training costs. The order of magnitude increase in computational power with Watermelon marks its transition from the Llama series to closed-source cutting-edge models, aiming to reshape the competitive landscape in consumer applications.
Similar to the computational arms race between OpenAI and Google, Meta is evolving from open-source distribution to high-performance closed models, where computational scale becomes a barrier, benefiting from data center expansion and supply chain optimization.
Structural assessment: This essentially represents capital concentration. Tech giants are concentrating resources on a few cutting-edge models through massive computational investments, creating a winner-takes-all scenario, making it difficult for smaller players to match training scales, thus increasing industry concentration.
ABAB News · Cognitive Law
Computational power is a barrier, creating generational gaps.
Those with abundant cash flow will dominate the pricing power of the next generation of models.
As competition intensifies, resources increasingly concentrate at the top.