Meta says its next-generation artificial intelligence model has reached the performance level of OpenAI's GPT-5.5 on internal benchmarks, signalling fresh momentum in the race among major AI developers to build more capable frontier models.
The claim was made by Alexandr Wang, Chief AI Officer at Meta Superintelligence Labs, during an internal company town hall, where he said the upcoming model, internally codenamed "Watermelon," has caught up with OpenAI's GPT-5.5 across key benchmark evaluations. The model is currently in training and is expected to succeed Meta's existing Muse Spark family of AI models.
According to Wang, Watermelon is being trained using significantly more computing power than its predecessor, reflecting Meta's continued investment in AI infrastructure. He also indicated that future updates would deliver notable improvements in coding and agentic capabilities, areas that have become central to competition among leading AI companies.
Meta has spent the past year strengthening its AI strategy under Chief Executive Officer Mark Zuckerberg, who has accelerated investments in data centres, AI chips and talent acquisition. The company recently reorganised its AI operations under the Meta Superintelligence Labs banner, bringing together research, product development and long term AI initiatives under a unified structure.
The announcement comes as frontier AI companies continue releasing increasingly capable large language models. OpenAI introduced GPT-5.5 earlier this year, positioning it as a major step forward in agentic reasoning, coding, scientific research and enterprise productivity. The model also improved performance across several industry benchmarks while maintaining similar response speeds to its predecessor.
Although Meta's claims are based on internal testing and have not yet been independently verified, they reflect the growing importance of benchmark performance in evaluating AI systems. Companies increasingly use standardised evaluations covering reasoning, coding, mathematics, knowledge retrieval and software engineering to demonstrate progress and attract enterprise customers.
Meta has faced increasing pressure to narrow the gap with OpenAI, Anthropic and Google, whose frontier models have gained strong adoption across consumer and enterprise markets. The company has responded by significantly expanding AI spending while recruiting senior researchers and engineers from across the industry. Meta has also raised its projected capital expenditure this year to support additional AI infrastructure and data centre capacity.
Industry analysts note that the competition has shifted beyond chatbot capabilities toward enterprise deployment, coding assistants, autonomous AI agents and multimodal systems capable of understanding text, images, audio and video. Improvements in reasoning accuracy, reliability and enterprise integration are increasingly becoming key differentiators for commercial AI platforms.
For marketers and enterprise technology teams, advances in frontier AI models are expected to influence a growing range of applications, including content creation, customer engagement, software development, marketing automation and business analytics. As vendors continue introducing more capable models, businesses are placing greater emphasis on evaluating performance, security, governance and integration capabilities before deployment.
Meta has not announced a public release date for Watermelon, and the company has not disclosed benchmark scores or technical specifications. However, Wang suggested that additional updates would arrive in the coming months as the company continues refining the model's capabilities.
The latest announcement underscores how rapidly the frontier AI landscape continues to evolve. As companies invest billions of dollars in computing infrastructure, specialised chips and AI research, competition is increasingly centred on delivering models that combine stronger reasoning, coding proficiency and enterprise readiness. Whether Meta's next model ultimately matches or surpasses its competitors will become clearer once independent evaluations and broader public releases begin.