What is the story about?
Meta's next flagship artificial intelligence model has reached a performance level comparable to OpenAI's GPT-5.5 on internal benchmark tests, Meta Superintelligence Labs chief Alexandr Wang reportedly told employees during a company-wide town hall, according to Business Insider, which cited people familiar with the meeting.
The model, internally codenamed Watermelon, is still in training, but Wang reportedly said it has already matched GPT-5.5 on several closely watched AI benchmarks. The publication noted that he did not specify which benchmarks were used to make the comparison.
If the assessment proves accurate, it would mark a significant milestone for Meta, which has spent the past year pouring billions of dollars into AI infrastructure, specialised chips and elite research talent in an effort to close the gap with rivals including OpenAI, Google and Anthropic.
According to Business Insider, Wang also gave employees fresh details about the scale of the project during the internal meeting.
"Watermelon, our next model after Avocado, is currently in training," Wang said, according to a person familiar with the meeting. "Watermelon uses an order of magnitude more compute than Avocado."
Avocado is Meta's internal codename for Muse Spark, the first model in a new family of AI systems that the company introduced in April. While Muse Spark performed well on several benchmarks, it was generally viewed as trailing the most capable models from OpenAI and Anthropic.
Wang also hinted publicly that Meta's AI models are improving rapidly. In a post on X on Thursday, he said an updated version of Muse Spark would launch soon with "major gains" in coding and agentic capabilities, describing it as another step towards narrowing the gap with competing AI systems.
When asked by an X user when Meta would have a coding model capable of matching Anthropic's Claude Opus, Wang replied, "Pretty soon," before adding that users would like what the company has "cooking."
Those comments come as Meta continues an aggressive recruitment drive led by CEO Mark Zuckerberg, who last year appointed Wang to lead the company's newly created Meta Superintelligence Labs.
Despite Wang's optimism, Zuckerberg reportedly struck a more cautious tone during the same town hall.
According to Reuters, the Meta chief executive told employees that AI agents had not "accelerated in the way" company executives had previously hoped.
Reuters also reported that Zuckerberg addressed the company's recent layoffs, describing them as not as "clean" as they should have been. He reportedly said the workforce reductions reflected concerns among senior leadership that Meta was not moving quickly enough to adapt to the rapidly evolving AI landscape.
The report added that Zuckerberg acknowledged the company's AI-focused organisational overhaul had yet to deliver its expected benefits. However, he reportedly told employees that he expects Meta's investments to begin producing more visible improvements over the next three to six months.
Meta has made AI its biggest strategic priority, backing that ambition with unprecedented spending. Reuters reported that the company could invest as much as $145 billion in AI infrastructure this year, alongside an aggressive campaign to hire leading AI researchers with compensation packages reportedly worth hundreds of millions of dollars.
The race has only intensified in recent months. OpenAI released GPT-5.5 in April and unveiled GPT-5.6 late last month, although the newer model has not yet been made generally available following requests from the US government.
For Meta, Watermelon is expected to be the clearest test yet of whether its heavy investment in computing power, infrastructure and talent can translate into AI models capable of competing with the industry's leading systems.
The model, internally codenamed Watermelon, is still in training, but Wang reportedly said it has already matched GPT-5.5 on several closely watched AI benchmarks. The publication noted that he did not specify which benchmarks were used to make the comparison.
If the assessment proves accurate, it would mark a significant milestone for Meta, which has spent the past year pouring billions of dollars into AI infrastructure, specialised chips and elite research talent in an effort to close the gap with rivals including OpenAI, Google and Anthropic.
Wang says Watermelon uses far more computing power
According to Business Insider, Wang also gave employees fresh details about the scale of the project during the internal meeting.
"Watermelon, our next model after Avocado, is currently in training," Wang said, according to a person familiar with the meeting. "Watermelon uses an order of magnitude more compute than Avocado."
Avocado is Meta's internal codename for Muse Spark, the first model in a new family of AI systems that the company introduced in April. While Muse Spark performed well on several benchmarks, it was generally viewed as trailing the most capable models from OpenAI and Anthropic.
Wang also hinted publicly that Meta's AI models are improving rapidly. In a post on X on Thursday, he said an updated version of Muse Spark would launch soon with "major gains" in coding and agentic capabilities, describing it as another step towards narrowing the gap with competing AI systems.
First, Mark was clearly talking about the industry’s progress on agentic capabilities on the whole.
But, while we’re on the topic: Our next Muse Spark update is coming soon. Big improvements in coding and agentic capabilities to be more competitive with other leading models.… https://t.co/uTjx8sZM2A
— Alexandr Wang (@alexandr_wang) July 3, 2026
When asked by an X user when Meta would have a coding model capable of matching Anthropic's Claude Opus, Wang replied, "Pretty soon," before adding that users would like what the company has "cooking."
Those comments come as Meta continues an aggressive recruitment drive led by CEO Mark Zuckerberg, who last year appointed Wang to lead the company's newly created Meta Superintelligence Labs.
Zuckerberg says AI agent progress has been slower than expected
Despite Wang's optimism, Zuckerberg reportedly struck a more cautious tone during the same town hall.
According to Reuters, the Meta chief executive told employees that AI agents had not "accelerated in the way" company executives had previously hoped.
Reuters also reported that Zuckerberg addressed the company's recent layoffs, describing them as not as "clean" as they should have been. He reportedly said the workforce reductions reflected concerns among senior leadership that Meta was not moving quickly enough to adapt to the rapidly evolving AI landscape.
The report added that Zuckerberg acknowledged the company's AI-focused organisational overhaul had yet to deliver its expected benefits. However, he reportedly told employees that he expects Meta's investments to begin producing more visible improvements over the next three to six months.
Meta has made AI its biggest strategic priority, backing that ambition with unprecedented spending. Reuters reported that the company could invest as much as $145 billion in AI infrastructure this year, alongside an aggressive campaign to hire leading AI researchers with compensation packages reportedly worth hundreds of millions of dollars.
The race has only intensified in recent months. OpenAI released GPT-5.5 in April and unveiled GPT-5.6 late last month, although the newer model has not yet been made generally available following requests from the US government.
For Meta, Watermelon is expected to be the clearest test yet of whether its heavy investment in computing power, infrastructure and talent can translate into AI models capable of competing with the industry's leading systems.
