Meta Pivoted to Proprietary AI Model Development
The company has announced the Watermelon model, signaling a departure from its previous open-weights strategy.
Updated on Sept. 23, 2026 in Artificial Intelligence
Live Poll
Do you trust proprietary AI models from big tech companies more than open-source alternatives?
Meta Platforms Chief AI Officer Alexandr Wang has announced the upcoming release of a new AI model codenamed Watermelon. This development represents a strategic shift toward proprietary software following the company's past focus on open-weight AI.
Why it matters
The change in strategy follows criticism regarding the 2025 release of Llama 4, prompting Meta to prioritize systems integrated into its own app and hardware ecosystem. This move marks a pivot away from the open-source approach that previously characterized the firm's AI trajectory.
The Watermelon model reportedly nears performance parity with OpenAI's GPT-5.5 by utilizing significantly increased computational resources. This follows the release of Muse Spark version 1.3 in September 2026.
The players
Meta Platforms
A technology company developing a stack of social apps, VR hardware, and large-scale AI models.
Alexandr Wang
The Chief AI Officer at Meta Platforms responsible for overseeing the company's current artificial intelligence research and product strategy.
The details
Meta is shifting its internal focus toward building models that are tightly woven into its existing app and hardware ecosystem. The performance gains in the upcoming Watermelon model are achieved by scaling up the computational resources allocated to training. This represents a transition from the model development methods used during the Llama 4 release period.
Timeline
April 2025: Meta released Llama 4.
June 2025: Alexandr Wang joined Meta.
April 2026: Meta launched the Muse Spark model.
September 2026: Meta released Muse Spark version 1.3.
The Tech Race
Meta's pursuit of performance parity with OpenAI's GPT-5.5 marks a significant shift in its AI roadmap. This transition follows the company's past reliance on open-weight releases like Llama 4 to challenge the industry's closed-model incumbents.
Users should expect future AI capabilities to be integrated directly into Meta's existing apps rather than distributed as standalone open-weight files. The specific hardware requirements and consumer-facing launch timeline for these features remain to be clarified.
The takeaway
Meta has moved away from the open-source philosophy that defined its previous AI cycles in favor of a closed, proprietary approach. Analysts and developers should monitor for future disclosures regarding the specific benchmark metrics for the Watermelon model to confirm its claimed parity.
Further reading
For more on the changing landscape of model development, see the Artificial Intelligence section.
Source note: This article includes information reported by Crypto Briefing.
Live Poll
Do you trust proprietary AI models from big tech companies more than open-source alternatives?
