Meta Launches Llama 4, Claims Breakthrough over GPT-4, Gemini

Samuel Mobolaji

Meta has taken a major leap in the artificial intelligence race with the introduction of its most advanced language models to date—Llama 4 Scout, Llama 4 Maverick, and the preview of the powerhouse Llama 4 Behemoth—setting a new standard in AI capabilities.

Over the weekend, CEO Mark Zuckerberg announced the rollout of these models, now powering Meta AI across WhatsApp, Instagram, Messenger, and meta.ai in more than 40 countries, including Nigeria, South Africa, Ghana, Uganda, Zambia, and Zimbabwe.

The newly released models, Scout and Maverick, are open-sourced and built for high performance across multimodal tasks. Llama 4 Scout is a 17-billion active parameter Mixture of Experts (MoE) model, featuring a record-setting 10 million-token context window and outpacing rivals like Gemma 3, Gemini 2.0 Flash-Lite, and Mistral 3.1 in industry benchmarks. Llama 4 Maverick, with 128 experts, boasts top-tier image grounding capabilities and outperforms GPT-4o and Gemini 2.0 Flash across several critical benchmarks. It achieves results comparable to DeepSeek v3 in reasoning and coding tasks—despite using half the active parameters—and delivers exceptional performance-to-cost efficiency, with a chat version scoring an impressive ELO of 1417 on LMArena.

These advancements are the result of distillation from Llama 4 Behemoth, Meta’s most powerful model in training, which is already showing benchmark dominance over GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro, especially in STEM-related evaluations. While Behemoth remains unreleased, its capabilities suggest Meta is fast closing the gap with the most advanced AI systems globally.

Meta says its focus is on building helpful and responsible AI. The Llama 4 models have been fine-tuned to deliver factual, non-judgmental answers, better follow instructions, and communicate with greater clarity and structure.

The rollout introduces a smarter and more responsive Meta AI that can answer questions, offer advice, generate ideas, and deliver well-formatted responses across a wide range of topics. Though multimodal features—like visual input—remain limited to users in the US for now, Meta confirmed plans to expand these capabilities globally within the year.

According to a Meta spokesperson, “Meta AI with Llama 4 is the assistant you can count on to provide helpful, factual responses without judgment. It’s more conversational, more precise, and more responsive than ever before.”

With the release of Scout and Maverick and a preview of Behemoth, Meta has positioned itself at the forefront of the next wave of generative AI. As Llama 4 rolls out globally, Meta is making it clear: the future of intelligent, accessible AI is already here—and it’s only just getting started.