Meta Platforms is preparing to launch its first AI models developed under Alexandr Wang. The company plans to offer some versions of these models under an open-source license, according to a report by Axios.

The rollout is expected to follow a staged approach. While some versions may be made publicly available, certain components will remain proprietary in the initial phase as the company assesses safety risks and safeguards more advanced capabilities.

Meta Platforms has been one of the few major U.S. tech firms to let developers modify its frontier models, but rising competition in artificial intelligence has led to growing expectations that it may reduce that level of openness.

Meta argues that its strength lies in its consumer reach. By integrating AI tools across platforms such as WhatsApp, Facebook, and Instagram, the company can deliver its technology to billions of users globally, often without direct cost, a scale that remains difficult for rivals to replicate.

The upcoming models are also part of an effort to close the gap with competitors. Meta’s earlier Llama 4 family lagged on several benchmarks, raising expectations for the next generation. According to Axios, the company does not expect to outperform rivals across every metric but believes it can differentiate in areas that resonate with everyday users.

Wang’s influence is increasingly visible in this direction. He has argued that Meta can help “democratize access” to advanced AI by offering tools that are widely available to developers and consumers. In contrast, competitors such as OpenAI and Anthropic are seen as focusing more on enterprise and government deployments with limited open access.

Balancing openness with control

Meta’s strategy is starting to come into focus as a hybrid approach. The company is looking to stay open enough to draw in developers, while keeping its most advanced systems closed to protect its competitive position.

The approach aligns with a wider industry shift. Even firms that once promoted open access are becoming more selective about releasing their most advanced models.

At the same time, tensions around openness have picked up. Elon Musk has criticized Sam Altman and OpenAI, arguing that the company has moved away from freely accessible models.

Meanwhile, Alibaba has chosen to keep its latest Qwen models proprietary after reversing its earlier open-source stance.

The developments come as debate intensifies within the AI community over the capabilities of current systems. Some researchers argue that models built on large-scale pattern recognition still fall short of genuine reasoning or human-like understanding.

Meta is also exploring alternative approaches alongside its core model development. One of those efforts is its “Brain Decoding” project, first previewed in 2023. The initiative focuses on understanding and simulating neural activity. It points to attempts to move beyond systems that mainly generate outputs from learned data patterns.