Key Highlights
- Alexandr Wang, Meta’s superintelligence chief, informed staff that the company’s forthcoming Watermelon AI model now rivals OpenAI’s GPT-5.5 in performance.
- The Watermelon model utilizes approximately ten times more computational power than Avocado, Meta’s codename for the Muse Spark model launched recently.
- Shares of Meta declined 4.90% after news of the internal announcement emerged, contradicting the optimistic technology claims.
- The company has revised its 2026 AI infrastructure budget to $125–$145 billion, representing an increase from the previous $115–$135 billion estimate.
- Wang hinted at substantial improvements in coding abilities and agentic functions coming to an updated version of Muse Spark.
Shares of Meta (META) fell 4.90% on Friday, despite positive signals about the company’s artificial intelligence development trajectory. During an employee town hall meeting, Alexandr Wang, who leads Meta’s superintelligence division, revealed that the company’s in-development model, internally called Watermelon, has achieved performance parity with OpenAI’s leading GPT-5.5 model — a development that could represent a significant breakthrough in Meta’s competitive AI strategy.
According to Wang, AI model benchmark testing supports this assertion, though the specific benchmarks used remain undisclosed.
Watermelon represents Meta’s successor to Avocado — the company’s internal designation for Muse Spark, unveiled in April. While Muse Spark demonstrated competitive benchmark performance upon release, it remained behind leading models from OpenAI and Anthropic.
Wang indicated that Watermelon leverages roughly ten times the computational resources compared to Avocado. Increased compute power generally correlates with enhanced model capabilities, though it simultaneously drives up operational expenses.
Enhanced Coding and Agentic Features Coming
Wang also shared updates publicly. Through a post on X, he announced an imminent update to the existing Muse Spark model, promising substantial enhancements in coding proficiency and agentic functionality.
Responding to a user inquiry about when Meta would deliver a coding model comparable to Anthropic’s Claude Opus, Wang indicated it would happen “pretty soon” and suggested users would appreciate what the company is developing.
Meta has invested extensively over recent years attempting to narrow the performance gap with OpenAI, Google, and Anthropic. Substantial capital allocation toward processing chips, data infrastructure, and elite AI researchers has yet to produce clear market dominance — at least not in publicly available offerings.
Zuckerberg brought Wang aboard last year to direct the newly branded Meta Superintelligence Labs. Wang manages an elite research division internally referred to as TBD, in addition to overseeing the company’s comprehensive AI and hardware initiatives.
AI Infrastructure Budget Continues Expanding
Meta has reportedly offered compensation packages worth hundreds of millions of dollars to attract premier AI researchers. The financial commitment extends far beyond talent acquisition.
The company informed investors that its projected 2026 expenditure on chips, data centers, and related infrastructure now ranges between $125 billion and $145 billion. This represents an upward revision from the earlier $115–$135 billion projection, attributed to escalating component prices and expanded data center construction.
For perspective, OpenAI launched GPT-5.5 in April. The organization subsequently introduced GPT-5.6 late last month — characterized as its most advanced model to date — though widespread access remains restricted, allegedly at the US government’s request.
Wang’s performance claims await independent verification, and Meta has not provided official commentary. OpenAI has not responded to requests for comment.
Get 3 Free Stock Ebooks
Discover top-performing stocks in AI, Crypto, and Technology with expert analysis.
- Top 10 AI Stocks - Leading AI companies
- Top 10 Crypto Stocks - Blockchain leaders
- Top 10 Tech Stocks - Tech giants
