Meta Claims Its Watermelon AI Model Has Caught Up With GPT-5.5

Meta’s Watermelon model is reportedly making major progress in the AI race, with superintelligence chief Alexandr Wang telling employees that the company’s next major system has caught up with OpenAI’s GPT-5.5 on key benchmarks.

According to Business Insider, Wang made the claim during an internal town hall. The model, codenamed Watermelon, remains in training and follows Avocado, Meta’s internal codename for Muse Spark.

Wang reportedly said Watermelon uses an order of magnitude more compute than Avocado. That suggests Meta has significantly scaled up its training efforts as it tries to close the gap with OpenAI, Google, and Anthropic.

Watermelon Could Be Meta’s Next Major AI Step

Watermelon appears to represent Meta’s next big push in frontier AI development. If Wang’s comments prove accurate, the model could mark a major improvement over Muse Spark and earlier internal systems.

The benchmark claim is notable because Meta has faced questions about whether its AI models can truly compete with the best systems from rival labs. Wang reportedly said Watermelon has reached GPT-5.5-level performance, though the exact benchmarks he cited remain unclear.

That detail matters. AI benchmark results can vary widely depending on the test, model configuration, and evaluation method. Still, the internal message suggests Meta believes its latest training run has moved the company closer to the front of the race.

Muse Spark Update May Improve Coding And Agents

Wang also said publicly on X that a Muse Spark update is coming soon. The update should bring major improvements in coding and agentic capabilities.

That would be important for Meta’s broader AI strategy. Coding performance has become one of the most closely watched areas in frontier AI, especially as companies race to build models that can write software, handle multi-step tasks, and operate more independently.

When asked when Meta would have a coding model on par with Anthropic’s Claude Opus, Wang said it would happen “pretty soon.” That suggests Meta sees coding and agentic AI as major priorities for Muse Spark and future models.

Meta Is Spending Heavily To Catch Rivals

Meta has invested heavily in chips, data centers, and AI talent. The company wants to compete directly with OpenAI, Google, and Anthropic at the top of the market.

However, Meta has struggled to convince developers and enterprise customers that its models lead the industry. Watermelon could help change that if it delivers the performance Wang described.

The reported progress also comes as Mark Zuckerberg continues to push an aggressive AI strategy. Zuckerberg has admitted that Meta’s AI restructuring has not gone as planned, but Watermelon may offer an early sign that the company’s investment is starting to produce stronger results.

OpenAI May Still Have A Lead

Even if Watermelon has caught up with GPT-5.5, Meta may still face a moving target. OpenAI reportedly debuted a stronger GPT-5.6 model late last month.

GPT-5.6 has not been generally released and is reportedly available only to government-approved partners. That means Meta may have narrowed the gap with GPT-5.5 while OpenAI already has a more advanced model waiting.

For now, Watermelon remains an internal model in training. Meta has not publicly released it, and the company has not confirmed when it could reach users or developers.