industry model-release ai-tools enterprise-ai openai

Meta Says Its Next AI Model Has Caught Up to GPT-5.5

Meta's Watermelon model matches GPT-5.5 on benchmarks, using 10x the compute of Muse Spark. Here's what we know.

AI Tutorials · 4 July 2026 · 2 min read

Quick answer

Meta's AI chief Alexandr Wang told employees that its next model, codenamed Watermelon, has matched OpenAI's GPT-5.5 on internal benchmarks. The model is still in training and uses 10x the compute of Meta's previous frontier model, Muse Spark. No release date has been announced.

industry

Meta Says Its Next AI Model Has Caught Up to GPT-5.5

Meta’s AI chief Alexandr Wang told employees this week that the company’s next frontier model, codenamed Watermelon, has matched OpenAI’s GPT-5.5 on internal benchmarks. The claim, first reported by Benzinga from an internal town hall, signals that Meta is closing the gap with the leading AI labs faster than many expected.

What We Know

Watermelon is the successor to Muse Spark, the model Meta released in April under the internal codename Avocado. Muse Spark performed well on standard benchmarks but did not match the top models from OpenAI or Anthropic. Watermelon appears to be Meta’s brute-force answer to that gap — Wang said it uses an order of magnitude more compute than its predecessor.

The model is still in training. Wang did not name the specific benchmarks where Watermelon matches GPT-5.5, and Meta has not published any results for independent verification. For now, this is an internal claim, not a proven result.

The Spending Behind It

Meta is backing these ambitions with enormous infrastructure investment. The company told investors it expects to spend between $125 billion and $145 billion this year on chips, data centres, and related AI infrastructure. That is up from an earlier guidance of $115 billion to $135 billion — and dwarfs what most competitors are spending.

Wang also hinted at a near-term update to Muse Spark with major improvements to coding and agentic capabilities. When an employee asked when Meta would ship a coding model on par with Anthropic’s Claude, Wang replied “pretty soon.”

What This Means for You

If Meta delivers on this claim, the implications are significant. Meta has a track record of open-sourcing its AI models through the Llama family, which means a GPT-5.5-class model could eventually become freely available to developers, researchers, and businesses. That would further compress the gap between proprietary and open-source AI.

For now, treat this as a signal worth watching rather than a settled result. Internal benchmarks from any company should be taken with caution until the model ships and independent evaluations confirm the claims. But if Watermelon lives up to the hype, it could shift the competitive landscape meaningfully — particularly if Meta follows its usual playbook and releases the weights publicly.

Frequently asked questions

What is Meta's Watermelon model?

Watermelon is the internal codename for Meta's next frontier AI model, currently in training at Meta Superintelligence Labs. It is the successor to Muse Spark (internally codenamed Avocado), which Meta released in April 2026.

How does Watermelon compare to GPT-5.5?

Meta's AI chief Alexandr Wang claims Watermelon matches GPT-5.5 on key benchmarks. However, Meta has not named the specific benchmarks used, and the results have not been independently verified.

When will Meta release the Watermelon model?

Meta has not announced a public release date. The model is still in training as of July 2026. Wang also mentioned that a near-term update to Muse Spark with improved coding and agentic capabilities is coming soon.

How much is Meta spending on AI infrastructure?

Meta told investors it expects to spend between $125 billion and $145 billion in 2026 on chips, data centres, and AI infrastructure — up from an earlier estimate of $115 billion to $135 billion.