Meta has unveiled its latest AI model, Llama 4, which is specifically designed to engage with more contentious topics, such as politics, compared to its predecessor, Llama 3.3.

Notably, Llama 4 has significantly reduced its refusal rate for politically or socially charged prompts from 7% to less than 2%, enhancing its responsiveness to hot-button questions.

CEO Mark Zuckerberg has committed up to $65 billion to AI projects this year, with aspirations for Llama to become the industry standard and potentially reach a billion users across Meta's platforms.

The Llama 4 suite includes three models: Scout, Maverick, and Behemoth, with Scout and Maverick officially released on April 5, 2025, while Behemoth remains in training.

Scout and Maverick are touted as Meta's most advanced models to date and are classified as open-weight models, allowing for some accessibility while keeping essential development details proprietary.

In a bid to eliminate bias, Meta acknowledges that many AI models have historically leaned left on contentious issues, and Llama 4 aims to present both sides fairly in its responses.

Testing has shown that Llama 4 only refuses to answer one side of a debated question 1% of the time, indicating a marked improvement in balance and fairness in its responses.

As of December 2024, Meta AI boasts a substantial user base, with 600 million users accessing its AI technologies each month.

The Llama 4 models are multimodal, capable of processing and integrating various forms of data, including text, video, images, and audio.