- Introduction to Meta’s technical achievement
- Details of the head of artificial intelligence’s statements
- Measurement standards and competition with the latest models
- Meta’s previous challenges and obstacles
- Massive investments in infrastructure and talent
- Frequently asked questions
Introduction to Meta’s technical achievement
In a new and striking escalation of the global technological arms race within the generative artificial intelligence sector, Meta recently announced to its employees that its upcoming software model, currently carrying the internal codename “Watermelon,” has finally managed to catch up with fierce competition and successfully matched the performance level of OpenAI’s advanced “GPT-5.5” model. This achievement reflects the new model’s ability to pass the industry’s most rigorous and closely watched benchmark and standard tests. These internal statements serve as the clearest and most explicit technical proof to date that Meta’s massive investments, valued at tens of billions of dollars in infrastructure and talent, are beginning to bear actual fruit in narrowing the wide technical gap with the industry’s leading innovators.
Details of the head of artificial intelligence’s statements
This encouraging news and information for the company’s future came from Alexandr Wang, who serves as Meta’s chief artificial intelligence officer, during a broad and comprehensive internal meeting involving all employees, according to a documented report published by the Business Insider platform. Wang clearly stated that the Watermelon model—the generative model succeeding their previous model known as “Avocado”—is currently undergoing intensive and complex training operations to ensure it reaches the required levels of efficiency and intelligence. Wang added, explaining: “The training process for the Watermelon model relies on and consumes computing power that multiplies and far exceeds that used in developing the Avocado model,” referring to the internal codename for the model officially introduced by the company to markets last April under the name “Muse Spark.” Despite these ambitious and encouraging claims, Wang did not precisely specify the nature of the benchmark tests or stringent standards relied upon to measure and compare the new model’s performance, and no independent technical verification supporting these claims has been provided at this time, while both Meta and OpenAI declined to provide any media or official comment in response to inquiries.
Measurement standards and competition with the latest models
It must be taken into account that the benchmark target referenced and relied upon by Wang in his evaluation, represented by the “GPT-5.5” model, pertains to a previous version of OpenAI’s models, which was announced and released last April. However, as definitive proof of the lightning speed and frantic pace at which this knowledge industry is evolving, OpenAI revealed late last month its most powerful and advanced model ever, the “GPT-5.6” model. Although this newest model has not yet been made available for commercial and public use in compliance with regulatory directives issued by the United States government, its very existence naturally means that even if the Watermelon model successfully matches the capabilities of the aforementioned version, Meta may still lag a significant step behind the new technical frontiers and horizons recently mapped out by its direct competitor.
Meta’s previous challenges and obstacles
The technical road has not been smooth for Meta in its competitive pursuit to develop advanced, reliable artificial intelligence systems; rather, its efforts have faced a series of repeated setbacks and obstacles. Its previous model, known as “Avocado,” was originally scheduled according to timelines for release in late 2025, but it suffered multiple technical delays after internal tests and evaluations showed clear shortcomings and weaknesses in logical reasoning skills, code writing, and text composition compared to competing models available on the market. When the model was finally launched after modifications in April under the name “Muse Spark,” it did not measure up to the leading and dominant models presented by OpenAI and Anthropic.
Massive investments in infrastructure and talent
Facing this existential challenge, CEO Mark Zuckerberg made exhaustive efforts and deployed immense financial and technical resources to bridge the gap his company suffers from in this vital and future-oriented arena. Meta expects its capital spending this year alone to range between $125 billion and $145 billion to secure advanced microchips, build giant data centers, and develop comprehensive technological infrastructure capable of accommodating these models. In mid-2025, Zuckerberg took a bold step by appointing Alexandr Wang—the former founder of Scale AI—to lead Meta’s super artificial intelligence labs, offering attractive and exceptional financial incentives valued at hundreds of millions of dollars per individual, with the aim of attracting the brightest minds and distinguished researchers in this sector to enhance the company’s innovative capabilities. The technological question will remain open as to whether the Watermelon model will truly fulfill its promises and benchmark ambitions, especially since the model is still in the training and development stages, and in an industry where the wheel of evolution never stops turning, finish lines and supremacy goals constantly keep moving forward.
Frequently asked questions
Question: What is the name of the new model Meta is developing?
Answer: Meta is currently training and developing an advanced language model carrying the internal codename “Watermelon.”
Question: How does Meta’s new model compare to competitors’ models?
Answer: The company’s head of artificial intelligence stated that the new model has reached a performance level matching the capabilities of the “GPT-5.5” model launched by OpenAI.
Question: What is the volume of investments allocated by Meta to the artificial intelligence sector this year?
Answer: The company’s spending is expected to range between $125 billion and $145 billion to develop the necessary chips, data centers, and technological infrastructures.
