Report says Meta's Watermelon model has reached GPT-5.5-level performance

Economic Times reports that Meta's upcoming AI model, code-named Watermelon, has reportedly reached performance levels comparable to OpenAI's GPT-5.5. The claim, attributed to Meta chief AI officer Alexandr Wang during an internal town hall, signals Meta's effort to regain credibility in frontier-model competition. The report follows months of pressure on Meta to improve coding, agentic capabilities, and internal AI infrastructure.

Key details: The model is code-named Watermelon, The reported performance comparison is to OpenAI's GPT-5.5, Alexandr Wang reportedly discussed the model during an internal town hall.

Why it matters: If the internal benchmark claim holds, Meta's model race may be tightening again after months of doubts about its frontier position.